Intrusion Detection Systems (IDS) face challenges due to high-dimensional network traffic and underrepresented rare attack classes. This paper presents SHAVA (SHAP-based Adaptive VANET Attack Detection), an open-source Python framework integrating explainable AI-driven feature selection with MASV-weighted SMOTE. SHAVA implements a modular pipeline for data validation, MASV feature quantification, percentile-based feature selection, and minority class augmentation. Evaluation on the CICIDS-2017 dataset demonstrates high detection performance across all classes with reduced feature dimensionality, while supporting real-time inference on desktop and edge devices representing Road Side Units (RSU) and On-Board Units (OBU) in vehicular networks. In addition to the methodology, SHAVA provides a reproducible, modular, and configurable software framework for experimental research, practical deployment, and adaptation to imbalanced learning scenarios, released under the MIT license.
Saharuna et al. (Tue,) studied this question.