AI-embedded IoT healthcare optimization with trust-aware mobile edge computing Authors: Malak Alamri, Khalid Haseeb, Mamoona Humayun, Menwa Alshammeri Published: 2025 ABSTRACT Embedded technologies combined with the Internet of Things (IoT), have transformed healthcare monitoring systems into automated and responsive platforms. In recent decades, many existing approaches have been based on edge computing to reduce response time in patient monitoring and provide a reliable method for interaction among the medical team and experts during disease diagnosis. Such approaches are the interconnection of battery-powered devices and physical objects to capture the physiological data streams for medical treatment and facilitate personalized healthcare systems. However, as wireless devices have limited resources for fulfilling end-user requests, this affects the accuracy of the medical system, especially in the presence of malicious devices on the communication infrastructure. Under diverse network conditions, such solutions lower the reliability level of the devices and increase the likelihood of suspicious processes. Therefore, to keep these significant concerns in IoT-based healthcare applications, trust and security should be adopted while collecting patients’ data over an insecure medium. In this research study, we propose a model referred to as Edge-Cloud Trusted Intelligence (ECTI), aiming to decrease the computing overhead on the devices. Additionally, multi-level security is implemented to ensure privacy preservation by adopting trusted behavior when communicating in a distributed environment. The edges utilize resources efficiently by employing task offloading strategies, enabling lightweight collaborative decision-making for routing in the healthcare domain. The performance results revealed notable improvement of the proposed model against related schemes in terms of various network metrics. Introduction Embedded healthcare applications have shown significant capabilities for real-time monitoring of patient treatment1,2. An edge network integrated with cloud computing enhances storage and processing capabilities, making communication channels less congested3–5. Such a system provides pre-processed data and continues to operate even in the event of disconnection or network interruption issues. Due to local processing at the edge, most health service providers provide rapid actions to medical experts for decision-making in emergency responses6,7. Many existing approaches leverage the intelligence of machine learning for data analytics and reduce resource consumption by utilizing minimal computational power in battery-powered devices8,9. The collaboration of distributed networks and sensor-based applications supports demanding and most critical operations in smart cities10,11. The hybrid nature of embedded modules not only processes the required system functionality on time but also enhances the performance in terms of responsiveness for health informatics12,13. Although, because of the openness of wireless systems and the interaction of heterogeneous devices, the chances of malicious threats lead to compromising the network infrastructure and increase the probabilities of data damage and breaches. Moreover, resources are often misused in responding to faulty requests; thus, security with a high level of trust is a significant research problem14,15. The joint interaction of data security and device trust enhances the communication paradigm’s reliability and facilitates adaptive decision-making while mitigating malicious behavior16,17. To address massive IoT data, it is filtered and preprocessed at the network edges, and transmission links are intelligently utilized to fulfill the demands of medical users for data analysis18,19. However, most of the proposed solutions explored edge computing with the support of cloud processing functionalities. However, the timely response, efficient utilization of network resources, along with security and trust, remain unsolved for complex computational applications, leading to threats and denying crucial healthcare services in a fault-tolerant manner20–22. Table 1 outlines the research challenges for IoT-based healthcare systems and the improvement of the proposed model against them.Table 1Challenges and Strategies of the Proposed IoT Healthcare ModelChallengesProposed strategiesThe increasing vulnerability of IoT devices and edge servers poses a risk to the integrity of healthcare data.Trust-aware mechanisms to evaluate device reliability and secure data offloading.Inefficient routing and low-latency communication are hindering real-time monitoring.AISET-Edge optimized routing to reduce latency and ensure timely healthcare data delivery.Adaptive traffic management inefficiencies across network layers.Intelligent traffic-aware decision-making for task offloading and load balancing.Inadequate security measures at edge servers are complicating safe deployment.Multi-layered security with anomaly detection (Random Forest) and lightweight encryption at edge nodes.Integration challenges of edge and cloud computing with IoT devices, balancing security, real-time analysis, and energy efficiency.Energy-efficient offloading and processing using trust scores and latency-aware selection of edge/cloud nodes. Table 1Challenges and Strategies of the Proposed IoT Healthcare ModelChallengesProposed strategiesThe increasing vulnerability of IoT devices and edge servers poses a risk to the integrity of healthcare data.Trust-aware mechanisms to evaluate device reliability and secure data offloading.Inefficient routing and low-latency communication are hindering real-time monitoring.AISET-Edge optimized routing to reduce latency and ensure timely healthcare data delivery.Adaptive traffic management inefficiencies across network layers.Intelligent traffic-aware decision-making for task offloading and load balancing.Inadequate security measures at edge servers are complicating safe deployment.Multi-layered security with anomaly detection (Random Forest) and lightweight encryption at edge nodes.Integration challenges of edge and cloud computing with IoT devices, balancing security, real-time analysis, and energy efficiency.Energy-efficient offloading and processing using trust scores and latency-aware selection of edge/cloud nodes. Challenges and Strategies of the Proposed IoT Healthcare Model Based on the aforementioned research problems, we highlight the contributions of the proposed model as follows. i.Develop a hybrid edge-cloud intelligence for IoT-enabled healthcare, providing efficient resource usage with latency-aware communication, and enabling scalable processing for timely decision-making and continuous monitoring.ii.Design and implement a trust-aware mechanism to assess device reputation, ensure patient data consistency, and maintain secure, reliable exchange of medical records.iii.Strengthen data security and integrity through lightweight encryption, anomaly detection, and cloud-assisted threat intelligence to counter cyber-attacks and ensure authentic health data.The remainder of this research study is as follows: Related work and formulation of problem statement are presented in Section “Related work”. Section “Proposed methodology” explains the working flow of the proposed model. In Section “Results”, experimental analysis is discussed. In the end, the conclusion and future work are outlined in Section “Conclusion”. i.Develop a hybrid edge-cloud intelligence for IoT-enabled healthcare, providing efficient resource usage with latency-aware communication, and enabling scalable processing for timely decision-making and continuous monitoring. Develop a hybrid edge-cloud intelligence for IoT-enabled healthcare, providing efficient resource usage with latency-aware communication, and enabling scalable processing for timely decision-making and continuous monitoring. ii.Design and implement a trust-aware mechanism to assess device reputation, ensure patient data consistency, and maintain secure, reliable exchange of medical records. Design and implement a trust-aware mechanism to assess device reputation, ensure patient data consistency, and maintain secure, reliable exchange of medical records. iii.Strengthen data security and integrity through lightweight encryption, anomaly detection, and cloud-assisted threat intelligence to counter cyber-attacks and ensure authentic health data. Strengthen data security and integrity through lightweight encryption, anomaly detection, and cloud-assisted threat intelligence to counter cyber-attacks and ensure authentic health data. Related work In modern healthcare systems, distributed IoT sensors enable continuous monitoring of patient vitals across diverse environments, from hospitals to remote care settings23,24. Healthcare applications are integrated with edge and cloud computing to develop a real-time decision support system25,26. They can interact with remote users and medical experts with the support of wearable sensors to capture the patient conditions and transmit them to processing servers for timely diagnosis and personalized treatment27,28. The local processing of edges enhances the responsiveness of the critical system and maintain the data resilience in the event of threats by secured offloading techniques29,30. In31, authors explore both fuzzy logic systems and the Whale Optimization Algorithm (WOA) to introduce a secure protocol for healthcare systems. A optimal cluster heads are selected using a WOA-based clustering framework to increase the stability of communication. Additionally, the trustworthiness of devices is assessed by evaluating a fuzzy-enabled trust mechanism and ensuring reliable data transmission within the IoT network. According to the performance results, the proposed protocol enhanced the energy efficiency of the healthcare system while also improving data security. To achieve privacy and authentic access control, the study32 introduces an Attribute Trust-based Security (ATST) algorithm, which aims to secure the transmission among connected devices and provide a more reliable environment. It guarantees secure message encryption and explores user attributes to develop a trust computational model. The ATST allows only authentic and authorized devices to be included in data encryption and decryption using a dual trust mechanism. The authors33 introduce a cross-layer and Cryptography-based Secure Routing (CLCSR) protocol to attain data privacy and protect the network from attacks. It is divided into two main stages: Firstly, a secure clustering mechanism is produced, and secondly, lightweight cryptography methods are exploited for secure data transmission. Triple cross-layer parameters are used for the selection of optimal cluster heads, which increases the network lifetime with improved load balancing. A fuzzy logic-based secure hierarchical routing scheme using the firefly algorithm (FSRF) is proposed34, aiming to enhance the trust of the healthcare applications. The three main frameworks are constructed: a fuzzy trust framework, a firefly algorithm-based clustering framework, and an inter-cluster routing framework. Initially, trust is computed using fuzzy logic and improves the reliability of the communicating devices. Secondly, it protects the systems against routing attacks and reduces the additional energy consumption for the network operations. Lastly, based on the fitness function, which comprises trust level, residual energy, hop count, communication radius, and centrality parameters, an optimum set of nodes is selected for the competition of cluster head. The results analysis has shown a significant improvement of the FSRF for various performance metrics. To address this risk, authors35 propose Adaptive Epsilon Greedy Reinforcement Learning (AEGRL), a variant of the traditional Epsilon-Greedy method, designed to secure data in both static and dynamic environments efficiently. In36, authors explore mobile edge computing for data aggregation from faultless device and proposed a trust-aware routing mechanism (TARM) . Normal and fault devices are identified using trust computation and enhances the confidence of IoT system. It also make ise of modified gray wolf optimization (GWO) for the construction of cluster and later for particular cluster heads an appropriate trust values are determined. Authors37 investigate the application of network coding in conjunction with opportunistic routing to enhance energy efficiency in wireless IoT infrastructure, taking into account of link correlation. Existing routing mechanisms assume link independence, leading to estimation errors in transmission calculations. To address this, authors propose an intra-session network coding mechanism that mines link correlation, along with an innovative routing method and an algorithm for selecting an optimal forwarder set based on accurate transmission estimations. Based on related studies, Table 2 provides a summary of the challenges faced in IoT-enabled healthcare systems, particularly in terms of security, communication efficiency, and data management. It provides a concise overview of the proposed model for mitigating these issues in a distributed and secure healthcare environment.Table 2Summary of challenges in IoT healthcare systems and proposed solutions.Research challengeCauseProposed methodologyVulnerability of IoT devices to security attacksIncreased exposure to cyberattacks leading to potential data leakage, especially at edge servers.Integration of edge and cloud computing with embedded IoT medical devices. Dual-layer defense strategy with real-time trust assessment, anomaly detection, and preliminary encryption at the edge.Inefficient routing mechanisms in IoT networksTraditional routing assumes link independence, leading to estimation errors in transmission calculations, affecting energy efficiency.Development of optimized communication paths, minimizing energy consumption and improving routing efficiency.Lack of low-latency communication in IoT networksCurrent methods fail to deliver efficient solutions for high-demand, time-sensitive communication.Implementation of ultra-low-latency communication solutions with optimized resource utilization for real-time healthcare monitoring.Adaptive traffic management inefficiency in IoT osmotic environmentsLack of proper management of traffic across multiple network layers, leading to inefficiencies.Adoption of adaptive traffic management mechanisms that optimize traffic flow and reduce congestion across IoT networks.Data security at edge serversData processed and stored at edge servers can be easily compromised, making security a significant concern.Robust security measures, including encryption and trust verification, to ensure data provenance and integrity. Table 2Summary of challenges in IoT healthcare systems and proposed solutions.Research challengeCauseProposed methodologyVulnerability of IoT devices to security attacksIncreased exposure to cyberattacks leading to potential data leakage, especially at edge servers.Integration of edge and cloud computing with embedded IoT medical devices. Dual-layer defense strategy with real-time trust assessment, anomaly detection, and preliminary encryption at the edge.Inefficient routing mechanisms in IoT networksTraditional routing assumes link independence, leading to estimation errors in transmission calculations, affecting energy efficiency.Development of optimized communication paths, minimizing energy consumption and improving routing efficiency.Lack of low-latency communication in IoT networksCurrent methods fail to deliver efficient solutions for high-demand, time-sensitive communication.Implementation of ultra-low-latency communication solutions with optimized resource utilization for real-time healthcare monitoring.Adaptive traffic management inefficiency in IoT osmotic environmentsLack of proper management of traffic across multiple network layers, leading to inefficiencies.Adoption of adaptive traffic management mechanisms that optimize traffic flow and reduce congestion across IoT networks.Data security at edge serversData processed and stored at edge servers can be easily compromised, making security a significant concern.Robust security measures, including encryption and trust verification, to ensure data provenance and integrity. Summary of challenges in IoT healthcare systems and proposed solutions. Proposed methodology In this section, we explain the discussion of our proposed model for an IoT-based healthcare system. It comprised AI-embedded decision-making for IoT routing, trust-aware strategies, and machine learning-based anomaly detection. The system model consists of sensors, edges, and communication links; thus, the consumption of network resources can be defined as given in Eq. (1).1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} R_{\text {total}} = \sum _{i=1}^{n} r_{s_i} + \sum _{j=1}^{m} r_{e_j}, \quad i \in \{1, \dots , n\}, \; j \in \{1, \dots , m\} \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r_{s_i}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r_{e_j}$$\end{document} denote the resource capacity for sensor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s_i$$\end{document} and edge node \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e_j$$\end{document}. Total number of sensors and edge nodes is represented by n, and the set of edge nodes is denoted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E = \{e_1, \dots , e_m\}$$\end{document}, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_{ij}$$\end{document} is the wireless channel for sensor i and edge j. The trust computation starts with the evaluation of direct trust based on the outcome of interactions between devices. Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_s(i,j,t)$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_f(i,j,t)$$\end{document} denote the number of successful and failed interactions between device i and device j at time t. The direct trust score \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{dir}_{ij}(t)$$\end{document} is computed using Eq. (2).2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} T^{dir}_{ij}(t) = \frac{I_s(i,j,t)}{I_s(i,j,t) + I_f(i,j,t)} \end{aligned}$$\end{document}where\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_s(i,j,t)$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_f(i,j,t)$$\end{document}: number of successful and failed interactions between devices i and j at time t\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{dir}_{ij}(t)$$\end{document}: direct trust score of device j from the perspective of device i at time ti, j: device indices, t: time instantIn the proposed model, the indirect trust score \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{ind}_{ij}(t)$$\end{document} is computed as given in Eq. (3).3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} T^{ind}_{ij}(t) = \frac{\sum _{k \in N_i} w_k \cdot T^{dir}_{kj}(t)}{\sum _{k \in N_i} w_k} \end{aligned}$$\end{document}where\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{ind}_{ij}(t)$$\end{document}: indirect trust score of device j from device i at time t\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_i$$\end{document}: set of neighbors of device i\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_k$$\end{document}: weight assigned to neighbor k based on its historical reliability\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{dir}_{kj}(t)$$\end{document}: direct trust of neighbor k on device j at time ti, j, k: device indices, and t: time instantThe overall trust score \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{ij}(t)$$\end{document} combines direct and indirect trust with tunable coefficients \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta$$\end{document} and is formulated using Eq. (4).4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} T_{ij}(t) = \alpha \cdot T^{dir}_{ij}(t) + \beta \cdot T^{ind}_{ij}(t), \quad \alpha + \beta = 1 \end{aligned}$$\end{document}Nodes with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{ij}(t)$$\end{document} below a predefined threshold \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta$$\end{document} are flagged for restricted participation in data forwarding or aggregation. To account for temporal variations, a time-decay factor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document} is applied to the trust score, giving more weight to recent behavior, as shown in Eq. (5).5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} T_{ij}(t) = \lambda \cdot T_{ij}(t-1) + (1 - \lambda ) \cdot T^{new}_{ij}(t) \end{aligned}$$\end{document}where trust determined from the updated interactions is denoted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{new}_{ij}(t)$$\end{document}. The edge nodes enhance trust evaluation through machine learning-based anomaly detection. Features extracted from sensor and interaction data include direct trust \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{dir}_{ij}$$\end{document}, indirect trust \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{ind}_{ij}$$\end{document}, packet delivery ratio, data consistency, and historical anomaly flags. A Random Forest classifier \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f_\theta$$\end{document} is trained to predict the probability that a device behaves maliciously using Eq.(6).6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} p_{anom}(i) = f_\theta (x_i) \end{aligned}$$\end{document}where the feature vector for device i is denoted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$x_i$$\end{document}, and later to obtain a trust-adjusted ML score, it is combined with the derived trust score, as given in Eq. (7).7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} T^{ML}_{ij}(t) = T_{ij}(t) \cdot (1 - p_{anom}(i)) \end{aligned}$$\end{document}Accordingly, the less trusted devices have given lower priority to be included in network-wide decision making, thus enhancing the reliability and robustness for trust computation. Equation (8) determines the final reputation score \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_j(t)$$\end{document} of a node j at time t across multiple edge nodes is aggregated using a weighted averaging mechanism, as shown in Eq.(8).8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} R_j(t) = \frac{\sum _{i=1}^{n} w_i \cdot T^{ML}_{ij}(t)}{\sum _{i=1}^{n} w_i} \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{ML}_{ij}(t)$$\end{document} denotes trust score of device i towards j from ML-based evaluation and reliability weight of device i, n is represented by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_i$$\end{document}. Using Eqs. (9) and (10), an adaptive encryption strength \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E_{ij}(t)$$\end{document} is applied based on the trust score to maintain data security.9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Delta E_{ij}(t)&= 1 - T^{ML}_{ij}(t) \end{aligned}$$\end{document}10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} E_{ij}(t)&= E_{\min } + (E_{\max } - E_{\min }) \cdot \Delta E_{ij}(t) \end{aligned}$$\end{document}Edge device continuously evaluates the malicious threats based on the direct and indirect trust, and ML-based anomaly predictions with temporal decay updates the reputation score. A stronger encryption technique is applied to lower-trusted devices to secure routing and mitigate network threats in a dynamic healthcare application. By combining direct trust, indirect trust from neighbors, ML-based anomaly probability, and a temporal decay factor, Eq. (11) updates the trust score \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{updated}_{ij}(t)$$\end{document} for node j from node i. Later, the adaptive encryption strength \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{enc}(i,j)$$\end{document} is determined by integrating the updated trust, resulting in stronger encryption for low-trusted devices, as defined in Eq. (12).11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} T^{updated}_{ij}(t)= & \lambda \Big [ \alpha \, T^{dir}_{ij}(t) + \beta \sum _{k \in N_i} w_k T_{ik}(t) \Big ] \cdot \big (1 - p_{\text {anom}}(j)\big ) \end{aligned}$$\end{document}12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} S_{enc}(i,j)= & S_{\min } + (S_{\max } - S_{\min }) \cdot \big (1 - T^{updated}_{ij}(t)\big ) \end{aligned}$$\end{document}where\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{dir}_{ij}(t)$$\end{document}: direct trust of node j as evaluated by node i, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{ik}(t)$$\end{document}: trust contribution from neighbor k\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_i$$\end{document}: set of neighbors of node i, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_k$$\end{document}: weight of neighbor k based on reliability\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha , \beta$$\end{document}: trust combination coefficients, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document}: temporal decay factor\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_{\text {anom}}(j)$$\end{document}: ML-predicted anomaly probability for node j\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{enc}(i,j)$$\end{document}: adaptive encryption strength, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{\min }, S_{\max }$$\end{document}: minimum and maximum encryption levelsTo maintain a global trust ledger, trust and reputation metrics are aggregated by a cloud server to ensure a trusted and more accurate threat detection system. The trust score for each node \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i$$\end{document} is updated using Eq. (13).13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \frac{dR(i)}{dt} = \gamma \cdot \frac{dR_{prev}(i)}{dt} + (1 - \gamma ) \cdot \frac{d\overline{T^{ML}_i}}{dt} \end{aligned}$$\end{document}where the updated reputation of node \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i$$\end{document} is denoted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R(i)$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_{prev}(i)$$\end{document} represents the historical score, past reputation, and recent ML-based trust is balanced with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document} balances. In contrast, the average ML-derived trust from recent interactions is denoted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{T^{ML}_i}$$\end{document}. Algorithm 1 explores direct trust from interactions and indirect trust from neighbors to compute overall trust \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{ij}$$\end{document} for IoMT devices. It utilizes weighted parameters \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta$$\end{document} for uniform contribution, and a decay factor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document} adjusts scores over time to maintain up-to-date trust evaluations. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_s(i,j,t)$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_f(i,j,t)$$\end{document}: number of successful and failed interactions between devices i and j at time t \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_s(i,j,t)$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_f(i,j,t)$$\end{document}: number of successful and failed interactions between devices i and j at time t \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{dir}_{ij}(t)$$\end{document}: direct trust score of device j from the perspective of device i at time t \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{dir}_{ij}(t)$$\end{document}: direct trust score of device j from the perspective of device i at time t i, j: device indices, t: time instant i, j: device indices, t: time instant \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{ind}_{ij}(t)$$\end{document}: indirect trust score of device j from device i at time t \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{ind}_{ij}(t)$$\end{document}: indirect trust score of device j from device i at time t \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_i$$\end{document}: set of neighbors of device i \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_i$$\end{document}: set of neighbors of device i \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_k$$\end{document}: weight assigned to neighbor k based on its historical reliability \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_k$$\end{document}: weight assigned to neighbor k based on its historical reliability \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{dir}_{kj}(t)$$\end{document}: direct trust of neighbor k on device j at time t \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{dir}_{kj}(t)$$\end{document}: direct trust of neighbor k on device j at time t i, j, k: device indices, and t: time instant i, j, k: device indices, and t: time instant \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{dir}_{ij}(t)$$\end{document}: direct trust of node j as evaluated by node i, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{ik}(t)$$\end{document}: trust contribution from neighbor k \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{dir}_{ij}(t)$$\end{document}: direct trust of node j as evaluated by node i, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{ik}(t)$$\end{document}: trust contribution from neighbor k \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_i$$\end{document}: set of neighbors of node i, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_k$$\end{document}: weight of neighbor k based on reliability \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_i$$\end{document}: set of neighbors of node i, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_k$$\end{document}: weight of neighbor k based on reliability \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha , \beta$$\end{document}: trust combination coefficients, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document}: temporal decay factor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha , \beta$$\end{document}: trust combination coefficients, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document}: temporal decay factor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_{\text {anom}}(j)$$\end{document}: ML-predicted anomaly probability for node j \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_{\text {anom}}(j)$$\end{document}: ML-predicted anomaly probability for node j \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{enc}(i,j)$$\end{document}: adaptive encryption strength, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{\min }, S_{\max }$$\end{document}: minimum and maximum encryption levels \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{enc}(i,j)$$\end{document}: adaptive encryption strength, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{\min }, S_{\max }$$\end{document}: minimum and maximum encryption levels Algorithm 1Reputation-Based Trust Computation and Lightweight Encryption for Edge-Enabled Healthcare Devices Reputation-Based Trust Computation and Lightweight Encryption for Edge-Enabled Healthcare Devices Algorithm 2 adjusts the trust scores \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{ij}$$\end{document} of health devices based on machine learning-based anomaly detection. For each device, the feature vector \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_j$$\end{document} is input into a trained Random Forest model to predict the anomaly probability \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_{anom}(j)$$\end{document}. The trust score \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{ij}$$\end{document} is updated by multiplying it with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(1 - p_{anom}(j))$$\end{document}. If the updated trust score \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{ML}_{ij}$$\end{document} falls below a predefined threshold \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta$$\end{document}, the device is marked as untrusted and excluded from routing or data aggregation; otherwise, it is marked as trusted for secure data forwarding. Algorithm 3 outlines the trusted communication and update the reputation of the IoT devices in healthcare system. The trust levels are adjusted adaptive using various attributes to mitigate the network threats against network anomalies. Algorithm 2ML-Based Anomaly Detection and Trust Adjustment between Health Devices ML-Based Anomaly Detection and Trust Adjustment between Health Devices In Fig. 1, the procedural flow of the ECTI model is presented by initializing the network and retrieving the interaction history for each neighbor. Later, direct and indirect trust are computed based on the feedback from neighbors and network conditions. It enhances the reliability in the healthcare environment and provides more consistent services to end-users. Moreover, a decay factor is applied to adjust trust scores, and if it is updated, a reevaluation is performed to achieve a more effective system. Otherwise, the algorithm proceeds to the next neighbor for transmitting the health record over the trustworthiness IoT-edge communication medium.Fig. 1Procedural flow of the proposed ECTI model for trust computation in healthcare environments. Procedural flow of the proposed ECTI model for trust computation in healthcare environments. Figure 2 illustrates the detection of anomalies using a predicted mechanism with trusted channels. It initializes and retrieves the feature vector for each health device, then trains the prediction model for anomaly detection using a trained Random Forest model. Later, the trust scores of the devices are reevaluated and adjusted based on the prediction level. Edges make an optimized decision by comparing the computed trust with the prefix threshold. The trust score, along with routing decisions, is integrated by the edges to ensure secure and optimal data forwarding. However, the device is excluded if the trust score is less than the threshold; otherwise, it is marked as trusted and included for secure data forwarding.Fig. 2Procedural flow of the ML-based anomaly detection and trust adjustment process. Procedural flow of the ML-based anomaly detection and trust adjustment process. Table 3 illustrates the various threat types and proposed methods, along with their descriptions, to achieve a trust-aware model. It provides a secure healthcare system and copes with malicious attacks, trust erosion, and anomalous behaviors within the system.Table 3Summary of Threat types and proposed trust-based mitigation strategies in IoT-healthcare.Threat typeProposed strategyDescriptionMalicious nodesDirect trust computation (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{dir}_{ij}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_s(i,j)$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_f(i,j)$$\end{document})The trust score for direct interactions is computed based on successful and failed interactions.Devices with consistently low scores are flagged as malicious.Reputation aggregation (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{ij}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta$$\end{document})The overall trust score is computed by combining direct and indirect trust from trusted neighbors.Low scores result in exclusion from secure operations.Trust erosionTrust Decay Factor (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document})A decay factor is applied to reduce the impact of outdated trust values, allowing the system to adapt to recent interactions.This ensures the system remains responsive to new behaviors.Adaptive Trust Updates (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{ij}(t)$$\end{document})Trust scores are dynamically updated to reflect recent data, ensuring long-term trust erosion does not compromise system performance.Anomalous behaviorsAnomaly Detection (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_{\text {anom}}(j)$$\end{document}, Machine Learning)Anomalous behaviors are detected using machine learning models that calculate the anomaly probability for each device.Trust is adjusted based on the likelihood of anomalous behavior.Trust Adjustment (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{ML}_{ij}$$\end{document})Devices with a high anomaly probability have their trust scores reduced, minimizing their impact on the system.Sybil attacksIndirect Trust Aggregation (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{ind}_{ij}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_k$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_i$$\end{document})Trust from multiple neighbors is aggregated to prevent a single malicious device from affecting the system’s trust calculations.It prevents Sybil attacks.Device exclusionTrust Threshold (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{\text {threshold}}$$\end{document})Devices are excluded from further operations if their trust score falls below a predefined threshold, ensuring that only trustworthy devices remain in the system.Continuous Trust Updates (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{ij}(t)$$\end{document})Trust scores are continuously monitored and updated. Devices with scores below the threshold are excluded from secure operations. Table 3Summary of Threat types and proposed trust-based mitigation strategies in IoT-healthcare.Threat typeProposed strategyDescriptionMalicious nodesDirect trust computation (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{dir}_{ij}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_s(i,j)$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_f(i,j)$$\end{document})The trust score for direct interactions is computed based on successful and failed interactions.Devices with consistently low scores are flagged as malicious.Reputation aggregation (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{ij}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta$$\end{document})The overall trust score is computed by combining direct and indirect trust from trusted neighbors.Low scores result in exclusion from secure operations.Trust erosionTrust Decay Factor (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document})A decay factor is applied to reduce the impact of outdated trust values, allowing the system to adapt to recent interactions.This ensures the system remains responsive to new behaviors.Adaptive Trust Updates (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{ij}(t)$$\end{document})Trust scores are dynamically updated to reflect recent data, ensuring long-term trust erosion does not compromise system performance.Anomalous behaviorsAnomaly Detection (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_{\text {anom}}(j)$$\end{document}, Machine Learning)Anomalous behaviors are detected using machine learning models that calculate the anomaly probability for each device.Trust is adjusted based on the likelihood of anomalous behavior.Trust Adjustment (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{ML}_{ij}$$\end{document})Devices with a high anomaly probability have their trust scores reduced, minimizing their impact on the system.Sybil attacksIndirect Trust Aggregation (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{ind}_{ij}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_k$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_i$$\end{document})Trust from multiple neighbors is aggregated to prevent a single malicious device from affecting the system’s trust calculations.It prevents Sybil attacks.Device exclusionTrust Threshold (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{\text {threshold}}$$\end{document})Devices are excluded from further operations if their trust score falls below a predefined threshold, ensuring that only trustworthy devices remain in the system.Continuous Trust Updates (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{ij}(t)$$\end{document})Trust scores are continuously monitored and updated. Devices with scores below the threshold are excluded from secure operations. Summary of Threat types and proposed trust-based mitigation strategies in IoT-healthcare. Algorithm 3Threat Model with Trust Management for Healthcare IoT System Threat Model with Trust Management for Healthcare IoT System Results This section provides the experimental results and their analysis using the NS-3 simulator. The proposed model is compared with AEGRL and TARM against varying health sensors and faulty devices. We deployed IoT health sensors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S = 100$$\end{document}–500 in the dimensional area of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$100~\text {m} \times 100~\text {m}$$\end{document}. The health sensors collect the patient data and transmit it to edge devices for local processing. The edges act as a central hub, forwarding incoming data to cloud servers after it has been authorized. The number of edge devices is set to 5, which are used for data gathering from health sensors and perform local processing. All devices of the same type have a predefined energy level of 5J. The transmission range varies from 10m to 50m. Table 4 declares the simulation parameters for conducting various experiments to analyze the proposed model.Table 4Simulation parameters.ParameterValueNumber of sensors (S)100–500Number of mobile edges (E)5Initial energy per sensor5 JCommunication range10–50 m\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta$$\end{document}0.5, 0.5Trust threshold (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta$$\end{document})0.6Decay factor (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document})0.8Simulation area100 m \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times$$\end{document} 100 mSimulation time500 sAnomaly injection rate5%Random Forest trees50 Table 4Simulation parameters.ParameterValueNumber of sensors (S)100–500Number of mobile edges (E)5Initial energy per sensor5 JCommunication range10–50 m\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta$$\end{document}0.5, 0.5Trust threshold (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta$$\end{document})0.6Decay factor (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document})0.8Simulation area100 m \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times$$\end{document} 100 mSimulation time500 sAnomaly injection rate5%Random Forest trees50 Simulation parameters. In our energy model, Eq. (14) defines the energy consumption of a node \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n \in {N}$$\end{document}.14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {E_n = \sum _{k} P_k \cdot t_k} \end{aligned}$$\end{document}where\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E_n$$\end{document}: total energy of node n\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_k$$\end{document}: Activity power k for sensing, transmission, reception, and processing\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_k$$\end{document}: time required to complete the activity kN: set of all nodes, including sensors and edge devicesFigure 3 illustrates that the performance of the ECTI model with related studies in terms of energy consumption based on varying numbers of IoT device scales. The statistical analysis demonstrated that the ECTI model outperforms existing solutions by an average of 52% and 57.8%. This improvement is due to the intelligence-driven routing mechanism, which explores a reliable and trust-aware ecosystem. It utilized edge-based decision-making to predict the next hop, along with link conditions such as congestion and load balancing, in a healthcare environment. Moreover, the integration of direct and indirect trust computation, combined with machine learning-based anomaly detection, enables the ECTI model to adapt and become more secure in forwarding health data. Moreover, the distributed processing model, along with its ability to evaluate past network behavior, ensures a lightweight mechanism that optimizes communication services and resource utilization. Timely identification and exclusion of malicious or faulty nodes prevent the propagation of false or redundant data, thereby reducing network overload and enhancing overall energy efficiency, particularly in scenarios with varying numbers of sensors. In Fig. 4, the performance of the ECTI model and existing approaches is compared for varying numbers of sensors in terms of network throughput. It was observed that the ECTI model effectively utilizes the communication links, increasing the network throughput by 33.4% and 39%, respectively. This performance is attributed to the intelligence of edges and the adaptation of trusted devices, which prioritize reliable communication in a critical system. In addition, the dual modes of trust enhance health-secure communication by ensuring reliable and secure interactions between IoT devices. It improves the detection of malicious attacks and prevents security breaches while ensuring the integrity of sensitive health records. Moreover, anomaly detection using machine learning techniques enables the real-time identification and exclusion of faulty devices, thereby reducing the amount of compromised data. Accordingly, the ECTI model contributes to higher throughput and guarantees the stability of network operations with less congested communication links. The use of a random forest for resource management reduces the network load, and the ECTI model dynamically adjusts routes by evaluating network conditions.Fig. 3Performance of proposed ECTI model with AEGRL, TARM for energy consumption with varying IoT sensors (100 to 500)Fig. 4Performance of proposed ECTI model with AEGRL, TARM for network throughput with varying IoT sensors (100 to 500) \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E_n$$\end{document}: total energy of node n \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E_n$$\end{document}: total energy of node n \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_k$$\end{document}: Activity power k for sensing, transmission, reception, and processing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_k$$\end{document}: Activity power k for sensing, transmission, reception, and processing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_k$$\end{document}: time required to complete the activity k \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_k$$\end{document}: time required to complete the activity k N: set of all nodes, including sensors and edge devices N: set of all nodes, including sensors and edge devices Performance of proposed ECTI model with AEGRL, TARM for energy consumption with varying IoT sensors (100 to 500) Performance of proposed ECTI model with AEGRL, TARM for network throughput with varying IoT sensors (100 to 500) In Fig. 5, the routing overhead is evaluated for the ECTI model and related solutions. Based on the performance results, it was observed that the ECTI model improved the routing overhead over existing approaches by an average of 30.4% and 37% due to less congested routes and reduced routing decisions and retransmission of healthcare data. This is achieved by dynamically updating data routes and only considering more reliable and trustworthy neighbors, resulting in a decrease in traffic load on IoT devices. In addition, it balances the load among the communication links by continuously assessing network conditions and adapting to the real-time conditions of the IoT environment. Figure 6 illustrates the proposed model’s performance in terms of trust stability, showing improvements of 30.5% and 39.4%, especially under faulty node conditions. This enhancement results from the integration of random forest-based trust prediction and trust decay mechanisms, which dynamically update trust scores based on network conditions. The model prioritizes reliable nodes and excludes faulty ones, ensuring optimal routing decisions. Additionally, reliable route optimization adjusts paths based on real-time trust values, ensuring fault tolerance and system reliability. It maintains high trust stability, even in the presence of compromised devices. The integration of data reliability increases trust levels by consistently demonstrating positive behavior of the devices and increases the consistency of their interactions. It also provides low failure rates for health data and improves network stability.Fig. 5Performance of Proposed ECTI model with AEGRL, TARM for routing overhead with varying IoT sensors (100 to 500)Fig. 6Performance of Proposed ECTI model with AEGRL, TARM for trust stability with varying IoT sensors (100 to 500) Performance of Proposed ECTI model with AEGRL, TARM for routing overhead with varying IoT sensors (100 to 500) Performance of Proposed ECTI model with AEGRL, TARM for trust stability with varying IoT sensors (100 to 500) The experimental results comparing the packet loss ratio of the ECTI model with existing solutions are shown in Fig. 7. As the number of IoT devices varies, the ECTI model demonstrates improvements in packet loss by 38.6% and 43.2%. This performance boost is due to the analysis of multiple criteria, combined with a machine learning algorithm, which enables optimal resource management and efficient handling of network congestion. By selecting key parameters based on real-time network conditions, the ECTI model ensures reliable communication links with trust-aware IoT data management. This results in more consistent next-hop decisions, providing long-term connectivity for embedded healthcare applications and ensuring effective performance for consistent devices. Fig. 8 further illustrates that the ECTI model outperforms existing solutions by 41.5% and 47%. This improvement is attributed to dynamic routing path updates, which identify nearby trustworthy neighbors and extend route lifetimes while reducing the traffic load on network devices. The optimization algorithm not only balances the load across peer data connections but also strengthens the trust-building process among devices. Additionally, routes that experience high data traffic are temporarily marked as untrusted, prompting the source node to formulate an alternative route. The ECTI model ensures high data delivery performance and enhances the utilization of network resources, even under dynamic conditions. Consequently, the proposed strategies enhance data management by achieving balanced traffic distribution and improved data reception in uncertain IoT wireless channels.Fig. 7Latency comparison of the proposed ECTI model with AEGRL and TARM under varying number of IoT sensors (100–500).Fig. 8Packet drop ratio of the proposed ECTI model compared with AEGRL and TARM for varying IoT sensor counts (100–500). Latency comparison of the proposed ECTI model with AEGRL and TARM under varying number of IoT sensors (100–500). Packet drop ratio of the proposed ECTI model compared with AEGRL and TARM for varying IoT sensor counts (100–500). Figure 9 illustrates the evaluation of the ECTI model against existing approaches for fault tolerance, and shows significant improvements by an average of 35.4% and 40% despite varying numbers of faulty nodes. It is due to the integration of trust computation, which provides a more consistent route selection for the transmission of real-time health data. It continuously adjusts the level of devices based on trust analysis computations with the network environment and communication parameters. The proposed ECTI model decreases the evaluation of unnecessary routing decisions and avoids the propagation of faulty data. By excluding unreliable or malicious nodes from the network, the trust computing of the ECTI model increases the level of fault tolerance. It offers a more trustworthy system by evaluating the behavior of IoT devices. In addition, the intelligence of edges ensures stable and secure communication even in the presence of faulty or compromised devices, ultimately preserving network stability and improving the overall efficiency of the IoT-based healthcare system. Figure 10 shows the performance of the ECTI model and the existing solution in terms of response time under varying faulty devices. The analysis, based on statistical outcomes, shows that the ECTI model improves response time by an average of 34% and 41% compared to existing solutions, due to trust-aware routing and updates that utilize a machine learning technique. Additionally, both direct and indirect trust evaluations exclude malicious devices, thereby enhancing the accuracy of anomaly detection policies. Furthermore, the optimization algorithm not only balances the load across peer data connections but also strengthens the trust-building process among devices. The route that experiences high data traffic is temporarily marked as invalid, prompting the source node to formulate an alternative route. The model ensures fault tolerance and enhances the utilization of network resources, even under dynamic conditions.Fig. 9Fault tolerance proposed ECTI model with AEGRL and TARM under varying fault rates.Fig. 10Analysis of response time for proposed ECTI model with AEGRL and TARM under varying fault rates. Fault tolerance proposed ECTI model with AEGRL and TARM under varying fault rates. Analysis of response time for proposed ECTI model with AEGRL and TARM under varying fault rates. The precision and recall metrics in Figs. 11 and 12 are crucial for evaluating the performance of our algorithm under faulty nodes. Precision measures the proportion of correctly identified nodes as functional out of all nodes identified as functional by the algorithm. On the other hand, recall evaluates the algorithm’s ability to identify all functional nodes, including those that are operational but may be incorrectly classified as faulty. In the presence of faulty nodes, a trade-off between precision and recall becomes evident, where improving one often comes at the expense of the other. To handle such trade-offs, our algorithm dynamically adjusts its decision thresholds based on network conditions, enhancing the robustness of the fault detection process. The precision-recall balance ensures that the system maintains high reliability and minimizes the occurrence of false positives and negatives in the identification of faulty nodes.Fig. 11Performance of Proposed ECTI model with AEGRL, TARM for recall with varying fault rate.Fig. 12Precision comparison of the proposed ECTI model with AEGRL and TARM under varying fault rates. Performance of Proposed ECTI model with AEGRL, TARM for recall with varying fault rate. Precision comparison of the proposed ECTI model with AEGRL and TARM under varying fault rates. Conclusion Intelligent techniques are utilized by many IoT devices for sensing and transmitting data in smart cities. These networks are integrated with sensors, digital services, and many processing sub-systems to communicate with the physical world. The integrated devices gather real-time data from the environment and forward it to remote users. Ultimately, easing the monitoring process with minimal human interference and facilitates the socialites. Although many approaches have been developed for smart healthcare, the majority of them are unable to cope with scalability and incur additional processing overheads for medical requests from IoT devices. Moreover, security is another significant research objective for many existing solutions. In this research, we propose a model utilizing edge–cloud interaction for Trust-aware communication in the healthcare industry. The intelligence is applied at the edge using an AI-driven approach to process the medical data locally and generate decisions with lower data latency. The trusts are computed mutually among devices to enhance the positive reputation for the transmission and provide a privacy-preserving environment against threats. Although the proposed model is limited to the availability of resources for optimizing healthcare. In future work, we aim to extend the overall performance of the proposed model by integrating SDN-based detection of distributed attacks and exploring some deep learning models to test the proposed model with realistic and large-scale datasets.