INTRUSION DETECTION SYSTEM FOR VEHICULAR ADHOC NETWORK USING DEEP LEARNING

Abstract

VANET's primary goal is to enhance safety, comfort, driving effectiveness, and reduce time spent in traffic congestion. However, it remains vulnerable to several security threats, including DoS, fuzzy, and impersonation attacks, due to its decentralized infrastructure. The absence of authentication information in the CAN bus, such as source and destination addresses, allows attackers to inject malicious messages easily, leading to severe system issues. This work proposes an RNN-based Deep Learning Intrusion Detection System (IDS) that applies clustering and classification methods to detect VANET intrusions using both LSTM and Simple RNN architectures. The offset ratio remains a critical parameter for intrusion detection, analyzing the gap between message requests and CAN responses. In 2024–2025, deep learning–driven IDS models have shown improved detection accuracy, leveraging temporal data and adaptive learning to identify multiple attack types effectively. The proposed RNN-IDS model offers a novel and efficient approach to enhancing intrusion detection precision and ensuring data integrity within modern vehicular networks.