SHAP-based feature selection and MASV-weighted SMOTE for enhanced attack detection in VANETs

Abstract

Vehicular Ad Hoc Networks (VANETs) are integral to Intelligent Transportation Systems (ITS) but remain highly vulnerable to cyberattacks, such as malicious attacks and position falsification. Detection is hindered by high-dimensional traffic data and severe class imbalance. Existing intrusion detection methods often overlook feature importance, limiting adaptability to different attack types. This study proposes an adaptive Intrusion Detection System (IDS) integrating SHAP-based feature selection with a MASV-weighted SMOTE technique. To the best of our knowledge, this is the first framework to leverage SHAP values not only for feature selection but also to guide class rebalancing during synthetic sample generation. Unlike conventional approaches, which treat all features equally, our method prioritizes features based on their Mean Absolute SHAP Values (MASV) in both selection and oversampling. Evaluated on CICIDS-2017 and validated on VeReMi, the framework demonstrates strong generalizability between datasets. It reduces feature dimensionality by up to 80% (78 to 15 features) while maintaining 99.91% accuracy, achieving up to 50.79% faster training and real-time inference below 0.1 ms per instance. MASV-weighted SMOTE transforms minority class detection performance, elevating the Infiltration attack F1-score from 0 to 88.89% and PR-AUC from 4.43% to 100%. These results outperform baseline models, enabling accurate, efficient, and interpretable IDS for VANETs security applications.