Advancing connected vehicle security through real-time sensor anomaly detection and recovery

Abstract

Connected Vehicles (CVs) are a crucial element in the evolution of smart transportation systems, utilizing communication and sensing technologies to interact with each other and with infrastructure. As these vehicles become more interconnected, the risk of their components being affected by anomalies or intentional malicious attacks grows. It is essential, therefore, to identify and filter out any anomalous data to ensure reliable decision-making. Existing solutions for anomaly detection in CVs include methods such as kalman filter, cumulative summation, convolutional neural networks and other machine learning models. However, a prevalent issue is the limited universality of anomaly datasets along with the variability introduced by simulated data. Additionally, there are few methods for recovering the network from anomalies using sensor information. In this paper, we address these limitations by utilizing the Tampa CV (TCV) dataset and incorporating anomalies such as bias, noise, and spikes. Furthermore, we present a novel method for real-time anomaly detection in CVs using Bayesian Online Change Point Detection (BOCPD). We propose a unique recovery mechanism that employs Bayesian forecasting to interpret identified anomalies, marking the first of its kind in this field. This approach significantly enhances the security of CV systems by seamlessly merging instant detection with swift recovery, ensuring continuous protection against data integrity threats. Results demonstrate that the proposed model achieves an average accuracy improvement of 53.83 % over other machine learning models. This paper makes advancement through real-time anomaly detection and recovery mechanisms, thus significantly improving the resilience of smart transportation systems against data integrity threats.