D-MNR: Dynamic Malevolent Node Recognition in VehicularCommunications — A Machine Learning Approach

Abstract

For Vehicular Ad Hoc Networks (VANETs) to be secure and private, it is essential to identify hostile activity. The problem of detecting black hole attacks, which seriously jeopardize the Dynamic Source Routing (DSR) protocol, is discussed in this study. While effective for static data, traditional machine learning-based approaches find it difficult to adjust to the constant stream of real-time data in VANETs. We use incremental classifiers, which can analyze data dynamically for prompt attack detection, to get around this issue. In particular, we assess how well Adaptive Random Forest (ARF) and K-Nearest Neighbours (KNN) identify black hole attacks. In order to capture important characteristics that define black hole behaviour, the data used for attack detection is simulated under controlled conditions. Key performance parameters, such as accuracy, precision, recall, F1-score, and training/testing timeframes, are used to evaluate the classifiers. Despite requiring longer training and testing durations, the results reveal that ARF performs better than KNN across all criteria, indicating greater detection capabilities for black hole attacks. The results emphasize the usefulness of ARF for enhancing security in dynamic network environments and demonstrate the potential of incremental online learning techniques for continuous and real-time threat detection in VANETs. This study suggests a scalable approach for real-time cyber security in automotive networks and provides fresh insights into improving threat detection techniques.