Intelligent IoT Attack Detection Design via ODLLM with Feature Ranking-based Knowledge Base

Satvik Verma, Qun Wang, E. Wes Bethel · 2025

Read Intelligent IoT Attack Detection Design via ODLLM with Feature Ranking-based Knowledge Base by Satvik Verma, Qun Wang, E. Wes Bethel (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-21674.txt.

Introduction

Satvik Verma, Qun Wang, and Wes Bethel wrote a 2025 paper on spotting attacks on internet of things networks. Small connected devices are now everywhere, and flood attacks that drown a target in traffic keep growing in scale and craft. Older machine learning tools often miss these attacks because the traffic mixes many patterns that keep changing.

Their design runs a large language model on the device itself, tuned for the task and backed by a knowledge base. Features of the traffic are ranked first, and two sizes of knowledge base match what small and large models can hold. Tests showed strong accuracy across attack types, above all with compact models at the network edge. The design also keeps data local, which helps privacy.

Catalog record

CREATORSatvik Verma, Qun Wang, E. Wes Bethel
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE36 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-21674

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Intelligent IoT Attack Detection Design via ODLLM with Feature Ranking-based Knowledge Base Satvik Verma, Qun Wang, E. Wes Bethel Abstract The widespread adoption of Internet of Things (IoT) devices has introduced significant cybersecurity challenges, particularly with the increasing frequency and sophistication of Distributed Denial of Service (DDoS) attacks. Traditional machine learning (ML) techniques often fall short in detecting such attacks due to the complexity of blended and evolving patterns. To address this, we propose a novel framework leveraging On-Device Large Language Models (ODLLMs) augmented with fine-tuning and knowledge base (KB) integration for intelligent IoT network attack detection. By implementing feature ranking techniques and constructing both long and short KBs tailored to model capacities, the proposed framework ensures efficient and accurate detection of DDoS attacks while overcoming computational and privacy limitations. Simulation results demonstrate that the optimized framework achieves superior accuracy across diverse attack types, especially when using compact models in edge computing environments. This work provides a scalable and secure solution for real-time IoT security, advancing the applicability of edge intelligence in cybersecurity. 1 Department of Computer Science, San Francisco State University, San Francisco, CA, 94132 2 Lawrence Berkeley National Laboratory, Berkeley CA, 94720 [email protected], [email protected], [email protected] Introduction The proliferation of IoT sensors in both residential and industrial environments has led to the generation of vast amounts of data that require timely and effective processing to facilitate rapid decision-making ( Zhang et al. 2023 ) . However, IoT sensors are particularly vulnerable to various cyber-attacks, especially DoS and Distributed DDoS attacks. These attacks can cause significant losses and further harm, making the quick and accurate identification of such threats critically important ( Jaton, Gyawali, and Qian 2023 ) . ML algorithms have been extensively used to detect abnormal DDoS traffic. The authors in ( Hussain et al. 2020 ) proposed a methodology to convert the network traffic data into image form and trained a CNN model for DDoS detection. ( Jia et al. 2020 ) presented a DDoS attack detection algorithm based on traffic variations and LSTM and CNN models. The authors in ( Aysa, Ibrahim, and Mohammed 2020 ) utilized LSVM, Neural Network, and Decision tree to detect abnormal activities such as DDOS features. Traditional ML algorithms rely on large datasets for training and face numerous limitations in this context. Moreover, when multiple attack types are mixed together, these algorithms often struggle to perform effectively. The emergence of large models has shown promise in the real-time and accurate identification of various abnormal network traffic patterns ( Zhu et al. 2024 ) . Nonetheless, these models typically require substantial computational resources for training and deployment, consuming significant amounts of computing power and electricity. Moreover, utilizing third-party models introduces data privacy and security concerns, which are particularly pertinent in network attack defense scenarios ( Yao et al. 2024 ) . In distributed, large-scale IoT networks, such models often fail to meet user needs promptly due to their resource-intensive nature and potential latency issues. This has led to increased interest in edge computing paradigms, where edge intelligence and on-device large models have garnered significant attention from researchers ( Xu et al. 2024a ) ( Chen, Li, and Ma 2024 ) . By employing techniques like model pruning and compression, smaller models can deliver functionalities comparable to their larger counterparts in most situations ( Xu et al. 2024a ) ( Chen et al. 2024 ) . Building upon this, developing applications on ODLLMs can ensure affordable intelligent decision-making and local data processing ( Xu et al. 2024b ) . However, when applying ODLLMs to network attack detection, a lack of necessary background knowledge within the models necessitates fine-tuning and the integration of knowledge base (KB) to enhance their performance. Therefore, we propose a novel system that leverages ODLLMs augmented with KB assistance to improve the detection accuracy of DDoS attacks in IoT environments. Our system addresses the challenges of computational resource constraints and privacy concerns by enabling on-device processing. We demonstrate that with appropriate KB support, ODLLMs can achieve performance comparable to larger models while operating within the limitations of edge devices. The contributions of this paper are as follows: A novel DDoS attack detection system that utilizes ODLLMs for IoT environments is proposed. We first introduce a novel approach for feature prioritization using a Random Forest Regressor (RFR) to rank the most critical features for different DDoS attack types, enabling the construction of efficient and scalable KBs. We then address the limitations of smaller ODLLMs by designing a simplified KB that retains only high-impact features, significantly improving predictive performance while reducing computational overhead. Third, we demonstrate the effectiveness of our framework through extensive experiments on the CICIoT 2023 dataset, achieving high accuracy in detecting various DDoS attack types. Our results highlight the critical role of tailored KB designs in enabling efficient and accurate attack detection on resource-constrained edge devices, paving the way for scalable and secure IoT network solutions. The subsequent sections are organized in the following manner. The system model and problem formulation are presented in Section II. The proposed feature ranking-based KB design is developed in Section III. The findings of the simulation are presented in Section IV. Finally, Section V provides the concluding remarks for this paper. Figure 1: System model. System Model and Problem Formulation System Model As shown in Fig. 1 , the framework operates in three primary stages: (1) Feature Analysis and Ranking: The system begins by extracting key features from the historical records of DDoS traffic data. These features include important values such as protocol types, packet rates, inter-arrival times, TCP flags, and statistical metrics like average packet size and variance. By analyzing these features, the system ranks them based on their significance in indicating abnormal network behavior. This prioritization allows the model to focus on the most impactful indicators of potential attacks. (2) Knowledge Base Construction: The KB serves as the intermediary layer between feature analysis and anomaly detection. It provides a structured repository of critical insights derived from the ranked features. The KB is constructed in two formats to optimize compatibility with different ODLLM model capacities. Long KBs are used for detailed analysis with medium-size ODLLM, and short KBs are used for lightweight ODLLM applications. The KB encapsulates the most distinguishing features of each attack type. These features are encoded as concise descriptors that facilitate quick comparison with incoming traffic data. (3) Integration with LLM for Anomaly Detection: After identifying the important features and constructing KB, the system integrates them with an ODLLM to perform anomaly detection. The LLM leverages its advanced reasoning capabilities and contextual understanding to interpret the feature set comprehensively. By incorporating domain-specific knowledge, the LLM can accurately predict the type of network attack occurring. By deploying the model on edge devices, we aim to maximize the accuracy of abnormal traffic detection. Types of DDoS Attack We consider four types of DDoS attacks and their characteristics: ICMP Flood Attack overwhelms the target with a high volume of ICMP echo requests, causing the network to become congested and unresponsive. This attack usually causes increased bandwidth consumption, degraded service performance, and potential downtime. UDP Flood Attack sends a large number of UDP packets to random ports on the target server, forcing it

Read the complete text