Constructing Optimal Sparse Decision Tree for Analying I-SIG System Attack

Xiaojin Wang, Yingxiao Xiang, Wenjia Niu, Endong Tong, Qi Alfred Chen, Gang Li, Jiqiang Liu, Long Li

Abstract: The Intelligent Traffic Signal System(I-SIG), has been deployed in states including New York, California and Florida by the US Department of Transportation, improving the traffic mobility, connectivity and efficiency in real-world intersections. Unfortunately, a new vulnerability to cause a congestion attack was revealed in 2018, which requires emerging security analysis and reinforcement. From high-dimension traffic features, to further evaluate attack consequences and analyze explainable feature influences, we propose an optimal sparse decision tree(OSDT)-based approach. Through massive feature engineering around attack performing, we finally define and determine 29 features, and collect 195 high-quality samples from VISSIM platform. This is the first standard dataset for statistical I-SIG analysis. Experiment shows that, facing random-deleted sparse traffic feature values, the OSDT can better explain the feature influence with a 78.6% prediction accuracy, compared to other state-of-art decision tree models.