Towards Robust AI Model for Cyber-Physical Intelligent Transportation Systems by Hash-Based Ensemble Learning
Kamrul Hasan, Varshini Guduru, Saleh Zein-Sabatto, Deo Chimba, Imtiaz Ahmed · 2023
In the evolving landscape of vehicular technology, autonomous and connected features lead to heightened connectivity among vehicles, intelligent devices, and infrastructures, broadening the vulnerability to cyber-attacks within the Internet of Vehicles (IoV) systems. Intrusion Detection Systems (IDSs) leverage Machine Learning (ML) to detect these malicious intrusions, but they face challenges from data poisoning attacks, given the critical role of training data in threat modeling. The vulnerability of training data emerges as a significant threat vector during ML model training. This study introduces a hash function-enabled ensemble ML training framework tailored for IDS, mitigating data poisoning attacks during model training. We test and validate our framework using hash-enabled ensemble ML algorithms-including random forests, support vector machines, and decision trees-on the benchmark CICIDS 2017 dataset. Results demonstrate that the hash-enabled random forests model guarantees a misclassification rate below 0.5%.