MACHINE LEARNING-BASED DETECTION OF CLIENT SELECTION ATTACKS IN WIRELESS NETWORKS

Abstract

Wireless communication networks have become the fundamental infrastructure supporting modern digital society by enabling ubiquitous connectivity among mobile devices, Internet of Things (IoT) platforms, industrial automation systems, healthcare applications, smart transportation, cloud computing, and intelligent cyber-physical environments. IEEE 802.11 wireless local area networks (WLANs) remain one of the most widely deployed communication technologies because of their flexibility, low deployment cost, and ease of integration. However, the open nature of wireless communication exposes these networks to numerous security threats that compromise confidentiality, integrity, availability, and network performance. Among these threats, client selection attacks have emerged as a sophisticated class of adversarial behavior in which malicious access points, rogue clients, or compromised network entities intentionally manipulate the client association process to attract, redirect, overload, or isolate wireless clients. Such attacks degrade communication quality, reduce throughput, increase packet loss, facilitate traffic interception, and create opportunities for additional attacks including denial-of-service, man-in-the-middle interception, traffic analysis, and credential theft. Traditional rule-based intrusion detection systems and threshold-driven monitoring mechanisms often fail to recognize these attacks because client association decisions continuously change according to mobility, signal strength, channel utilization, traffic demand, and environmental conditions. This paper proposes a machine learning-based detection framework for identifying client selection attacks in IEEE 802.11 wireless networks through intelligent behavioral analysis and real-time anomaly detection. The proposed framework continuously collects network observations including received signal strength, client mobility patterns, association requests, authentication responses, channel utilization, retransmission rate, beacon characteristics, packet delay, traffic load, session duration, roaming frequency, and medium access behavior. The collected information undergoes preprocessing, feature engineering, normalization, dimensionality reduction, and behavioral profiling before being analyzed using supervised and unsupervised machine learning algorithms capable of distinguishing legitimate client association behavior from malicious manipulation. The framework integrates statistical learning, ensemble classification, anomaly detection, and confidence-based decision fusion to improve attack detection accuracy while minimizing false alarms. Experimental evaluation under representative wireless networking scenarios demonstrates that the proposed framework achieves higher detection accuracy, improved precision and recall, reduced false-positive rates, and earlier identification of client selection attacks compared with conventional signature-based and threshold-oriented security mechanisms. The proposed architecture provides a scalable, adaptive, and intelligent security solution suitable for enterprise WLANs, campus networks, industrial wireless environments, smart cities, healthcare systems, and nextgeneration intelligent wireless infrastructures.