XAI-IDS: DeepLIFT-AE Driven Granular Feature Extraction for Intelligent Detection of Emerging Vehicular Network Threats

The growing reliance on the Internet of Vehicles (IoV) has intensified the need for robust Intrusion Detection Systems (IDS), as recent studies reveal that over 60% of vehicular network breaches result from DoS and spoofing attacks, contributing to annual losses exceeding $1.2 billion globally. Despite advancements, conventional IDS frameworks still suffer from limited interpretability, high false alarm rates, and inadequate adaptability to evolving attack patterns. To address these challenges, this research introduces a novel eXplainable Artificial Intelligence (XAI)-based IDS framework utilizing the CICIoV-2024 dataset, which comprehensively represents real-time vehicular communication including Benign, DoS, and Spoofing traffic. The data undergoes rigorous preprocessing to remove noise and standardize features. A hybrid Fuzzy Probability Clustering with SHapley Additive Explanations (FPC-SHAP) is employed for granular feature extraction by estimating uncertainty and relevance simultaneously. Subsequently, DeepLIFT-Autoencoder (DeepLIFT-AE) is applied for feature ranking, capturing both low- and high-level semantic patterns. For classification, a Local Interpretable Model-agnostic Explanations-based Globally Randomized Ensemble (LIME-GRE) model is implemented, enabling highly interpretable decision boundaries across the three attack classes. The proposed XAI-IDS method achieved excellent performance in the evaluation, with an Accuracy of 99.966% and a Precision of 99.987%. It also demonstrated strong results in Recall at 99.786% and an F1 Score of 99.886%, indicating highly effective classification, which are increased over state of art methodologies.