2021 IEEE Wireless Communications and Networking Conference (WCNC) | 2021
WSG-InV: Weighted State Graph Model for Intrusion Detection on In-Vehicle Network
Abstract
The paper presents WSG-InV, a novel weighted state graph (WSG) model for lightweight IDS on in-vehicle network. By capitalizing on historical in-vehicle data of timestamps, message identifiers, and data field, WSG-InV constructs offline a weighted state graph $\\mathcal{G}=(\\mathcal{V},\\mathcal{E})$ where distinct message identifiers constitute the set $\\mathcal{V}$ of vertices and the edges in $\\mathcal{E}$ define the time-varying state transitions of the CAN frames. The iconic constituents of given in-vehicle data are condensed into a collection of ordered triples (the vectorized weight) that are further assigned to the edges in $\\mathcal{E}$. In the mean time, several kinds of intrusion data are evoked and the random forest model is deployed to conduct intrusion classification. WSG-InV then segments the online data stream into a slice of sliding windows and extracts a weighted state subgraph $\\mathcal{S}$ for each of them. By consulting $\\mathcal{G}$ as a benchmarking as well as optimizing a particular 3-variable programming, WSG-InV assesses the subgraph $\\mathcal{S}$ and thereby recognizes the corresponding traffic as normal or anomaly. Besides, WSG-InV can distinguish which type of attack the anomaly gears toward. Experimental results demonstrate almost optimal performance.