Abstract
The Internet of Vehicles (IoV) may face challenging cybersecurity attacksthat may require sophisticated intrusion detection systems, necessitating arapid development and response system. This research investigates theperformance advantages of GPU-accelerated libraries (cuML) compared totraditional CPU-based implementations (scikit-learn), focusing on the speed andefficiency required for machine learning models used in IoV threat detectionenvironments. The comprehensive evaluations conducted employ four machinelearning approaches (Random Forest, KNN, Logistic Regression, XGBoost) acrossthree distinct IoV security datasets (OTIDS, GIDS, CICIoV2024). Our findingsdemonstrate that GPU-accelerated implementations dramatically improvedcomputational efficiency, with training times reduced by a factor of up to 159and prediction speeds accelerated by up to 95 times compared to traditional CPUprocessing, all while preserving detection accuracy. This remarkableperformance breakthrough empowers researchers and security specialists toharness GPU acceleration for creating faster, more effective threat detectionsystems that meet the urgent real-time security demands of today's connectedvehicle networks.