Intrusion detection on internet of vehicles via combining log-ratio oversampling, outlier detection and metric learning

In recent years, the Internet of vehicles (IoV) technology becomes a research hotspot. However, it also becomes a hotbed for malicious attacks. In the IoV, frequent data transmission and complex connections among numerous different nodes increase the complexity and diversity of malicious attacks. In order to realize the accurate and rapid detection of malicious attacks in the IoV environment, in this paper, an intrusion detection method is proposed by combining oversampling, outlier detection and metric learning. The proposed approach improves intrusion detection effect in three main ways: 1) it oversamples the minority classes based on a novel strategy, 2) it introduces a new feature with basis of imbalance ratio, and 3) it reduces the outliers and rescales original samples actively to make the decision boundary clearer by combining outlier detection and distance metric learning. Furthermore, genetic algorithm is used to extract the optimal subset of features. The experimental results show that the proposed method can achieve 98.51% accuracy and maintain 0.82% false alarm rate on UNSW-NB15 dataset. Also, it outperforms the existing methods on ROAD, Car-Hacking and CAN-intrusion dataset for in-vehicle communications.