Temporal-Spatial Attention Network (TSAN) for DoS Attack Detection in Network Traffic

Bisola Faith Kayode, Akinyemi Sadeeq Akintola, Oluwole Fagbohun, Egonna Anaesiuba, Onyekachukwu Ojumah, Oluwagbade Odimayo, Toyese Oloyede, Aniema Inyang, Teslim Kazeem, Habeeb Alli, Udodirim Ibem Offia, Prisca Chinazor Amajuoyi · 2025

Read Temporal-Spatial Attention Network (TSAN) for DoS Attack Detection in Network Traffic by Bisola Faith Kayode, Akinyemi Sadeeq Akintola, Oluwole Fagbohun, Egonna Anaesiuba, Onyekachukwu Ojumah, Oluwagbade Odimayo, Toyese Oloyede, Aniema Inyang, Teslim Kazeem, Habeeb Alli, Udodirim Ibem Offia, Prisca Chinazor Amajuoyi (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-2503-16047.txt.

Introduction

Bisola Faith Kayode and a team of co-authors set out a new network defense model, called TSAN, to spot denial-of-service attacks in network traffic. Their paper is from two thousand twenty-five.

Such attacks flood a service to knock it out, and they cost firms a great deal. Old rule-based and count-based tools adapt poorly as attack forms change. The new model watches both when traffic moves and how it spreads, so it can catch odd shapes that old tools miss. It joins a transformer-style encoder for time with a shaped filter for space. The aim is a guard that keeps pace with fast-changing attacks.

Catalog record

CREATORBisola Faith Kayode, Akinyemi Sadeeq Akintola, Oluwole Fagbohun, Egonna Anaesiuba, Onyekachukwu Ojumah, Oluwagbade Odimayo, Toyese Oloyede, Aniema Inyang, Teslim Kazeem, Habeeb Alli, Udodirim Ibem Offia, Prisca Chinazor Amajuoyi
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE51 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-16047

Browse the full library · About the Tower of Babel

Full text

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

Temporal-Spatial Attention Network (TSAN) for DoS Attack Detection in Network Traffic Bisola Faith Kayode, Akinyemi Sadeeq Akintola, Oluwole Fagbohun, Egonna Anaesiuba, Onyekachukwu Ojumah, Oluwagbade Odimayo, Toyese Oloyede, Aniema Inyang, Teslim Kazeem, Habeeb Alli, Udodirim Ibem Offia, Prisca Chinazor Amajuoyi Abstract Denial-of-Service (DoS) attacks remain a critical threat to network security, disrupting services and causing significant economic losses. Traditional detection methods, including statistical and rule-based models, struggle to adapt to evolving attack patterns. To address this challenge, we propose a novel Temporal-Spatial Attention Network (TSAN) architecture for detecting Denial of Service (DoS) attacks in network traffic. By leveraging both temporal and spatial features of network traffic, our approach captures complex traffic patterns and anomalies that traditional methods might miss. The TSAN model incorporates transformer-based temporal encoding, convolutional spatial encoding, and a cross-attention mechanism to fuse these complementary feature spaces. Additionally, we employ multi-task learning with auxiliary tasks to enhance the model’s robustness. Experimental results on the NSL-KDD dataset demonstrate that TSAN outperforms state-of-the-art models, achieving superior accuracy, precision, recall, and F1-score while maintaining computational efficiency for real-time deployment. The proposed architecture offers an optimal balance between detection accuracy and computational overhead, making it highly suitable for real-world network security applications. K eywords Denial of Service (DoS) Attacks \cdot Network Security \cdot Deep Learning \cdot Attention Mechanisms \cdot Temporal-Spatial Features \cdot Transformer Networks \cdot Multi-task Learning \cdot Cross-attention \cdot Network Traffic Analysis \cdot Intrusion Detection Systems \cdot NSL-KDD Dataset \cdot Real-time Detection. NOTE: Preprint, Submitted to ICMLT 2025, Helsinki, Finland Bisola Faith Kayode is an Independent Researcher based in the United Kingdom (e-mail: [email protected]). Akinyemi Sadeeq Akintola is with Universidade NOVA de Lisboa, Lisbon, Portugal (e-mail: [email protected]). Oluwole Fagbohun is with Carbonnote, USA (e-mail: [email protected]). Egonna Anaesiuba-Bristol is with the University of Surrey, United Kingdom (e-mail: [email protected]). Onyekachukwu Ojumah is an Independent Researcher based in the United Kingdom (e-mail: [email protected]). Oluwagbade Odimayo is with the University of Essex, United Kingdom (e-mail: [email protected]). Toyese Oloyede is an Independent Researcher based in the United Kingdom (e-mail: [email protected]). Aniema Inyang is with Vuhosi Limited, United Kingdom (e-mail: [email protected]). Teslim Kazeem is an Independent Researcher based in the United Kingdom (e-mail: [email protected]). Habeeb Alli is with Readrly Limited, United Kingdom (e-mail: [email protected]). Udodirim Ibem Offia is with Vuhosi Limited, United Kingdom (e-mail: [email protected]). Prisca Chinazor Amajuoyi is with Vuhosi Limited, United Kingdom (e-mail: [email protected]). 1 Introduction Network security remains a critical concern in today’s interconnected digital landscape, with Denial of Service (DoS) attacks representing one of the most prevalent and disruptive threats faced by organizations worldwide [ 1 ] . As depicted in Figure 1, these attacks aim to exhaust system resources and disrupt legitimate network services, causing significant operational disruptions and financial losses [ 2 ] . The increasing sophistication and frequency of such attacks necessitate advanced detection mechanisms that can identify attack patterns with high accuracy and minimal false alarms. Figure 1: : DoS Attack Detection Overview. Illustration of the DoS attack detection process showing attackers targeting a network and the TSAN model analyzing traffic to produce detection results. Traditional approaches to DoS detection have relied heavily on statistical analysis, signature-based methods, or shallow machine learning algorithms [ 3 , 4 , 5 ] . While these approaches have shown some success, they often struggle with novel attack variants and demonstrate limited capabilities in capturing the complex temporal and spatial relationships in network traffic data. Deep learning methodologies have emerged as promising alternatives, offering improved detection performance through their ability to automatically learn hierarchical feature representations from raw data without extensive feature engineering [ 5 , 6 , 7 ] Recent advances in attention mechanisms, particularly in the field of natural language processing and computer vision, have opened new avenues for network security applications. Attention allows models to focus on the most relevant parts of the input data, potentially enhancing their ability to detect subtle attack patterns [ 8 ] . Despite these advantages, few works have systematically explored how attention mechanisms can be adapted to the specific challenges of network traffic analysis, where data exhibits both temporal dependencies across time windows and spatial relationships within individual network packets [ 9 ] . In this paper, we introduce a novel Temporal-Spatial Attention Network (TSAN) for DoS attack detection. Unlike previous approaches that typically focus on either temporal or spatial aspects of network traffic, our model leverages both dimensions simultaneously [ 9 ] . The TSAN architecture incorporates a transformer-based temporal encoder to capture sequential patterns across traffic windows, a convolutional neural network (CNN) based spatial encoder to extract features from individual network packets, and a cross-attention mechanism to fuse these complementary feature spaces effectively [ 10 ] To enhance the model’s robustness and generalization capabilities, we adopt a multi-task learning framework where the primary task of DoS detection is supplemented with auxiliary tasks including traffic pattern prediction, protocol distribution modeling, and temporal consistency checking [ 11 ] . This approach encourages the model to learn more comprehensive and nuanced representations of network traffic. Additionally, we incorporate a self-supervised pretraining phase to initialize the model parameters, further improving its ability to extract meaningful features from the network data [ 12 ] . Extensive experiments on the widely-used NSL-KDD dataset demonstrate that our TSAN model achieves superior performance compared to state-of-the-art methods for \operatorname{DoS} attack detection [ 13 ] . Beyond the performance improvements, we provide in-depth analyses of the model’s attention patterns and feature importance, offering insights into how the model makes decisions and which network features contribute most significantly to attack detection [ 13 ] . These insights not only validate our approach but also provide valuable knowledge for network security practitioners in developing more effective defense strategies [ 14 ] . 2 Literature Review The field of network intrusion detection has evolved significantly from its early days of signature-based systems to the current landscape dominated by machine learning and deep learning approaches. This review examines the trajectory of this evolution, focusing on DoS attack detection methods and the incorporation of attention mechanisms in network security applications. As illustrated in Figure 3 , the development of DoS attack detection methods has progressed through distinct research streams over the past two decades. The timeline visualization demonstrates how traditional machine learning approaches from the early 2000s have given way to increasingly sophisticated deep learning models, attention mechanisms, and most recently, multi-task and self-supervised learning paradigms. Our proposed Temporal-Spatial Attention Network (TSAN) represents the culmination of these research advances, integrating temporal and spatial features with cross-attention mechanisms for improved DoS attack detection. Figure 2: Evolution of DoS Attack Detection Methods. Timeline showing progression from traditional machine learning (2000-2010) to deep learning (2010-2015), attention mechanisms (2015-2020), and multi-task learning (2018-present), with TSAN representing the integration of these advances. 2.1 Traditional Machine Learning Approaches Early efforts in applying machine learning to network intrusion detection largely focused on shallow algorithms. [ 15 ] provided one of the first comprehensive surveys of intrusion detection systems, categorizing approaches based on detection methods and behavioral modeling. Their work highlighted the potential of data-driven approaches but was limited by the computational capabilities of the time. Mukkamala et al. (2002) [ 16 ] compared Support Vector Machines (SVM) and Neural Networks for intrusion detection, finding

Read the complete text