Anomaly Detection: Under the [data] hood in Smart Cars

Abstract

This research focuses on discovering baseline models for driving behavior and vehicle functioning from data in smart cars. This facilitates detection of anomalous behaviors that deviate from such baselines. Human behavioral patterns capture frequent or repeated behaviors of users in the data. Here the data is from smart car sensors which capture driving behaviors as a direct function of how the vehicle responds to the use by drivers. We define models that represent these patterns in vehicle data to associate with the human behaviors for the detection of the anomalous situations in driving. We deal with different scales and resolutions of time where driver behavior is captured. We validate our findings of discovering behavioral baselines with frequent patterns discovered in the data. These computational models for driver behavior can provide baselines as well as help to discern truly adverse incidents. We can also apply these models to identify emerging cyber threats on smart cars.