Driver authentication by quantifying driving style using GPS only

Driver authentication that is verifying a driver's identity constitutes an important aspect of modern day automobile. A driver gets access to drive a car based on his identity. Identity is normally verified with smart card or possession of key. Also there exist identity verification based on fingerprint, passcode and image based approach using camera mounted inside a car. However such approach do not consider driving style based authentication. In this paper, we present an approach that uses personalized statistical feature set extracted from the global positioning system (GPS) data to authenticate a driver. Such personalized feature set reduces computation, improves interpretability of features and accuracy. Proposed method is further enriched by determining and using the most suitable machine learning technique. Our approach is tuned to increase accuracy, sensitivity and specificity. Overall mean area under receiver operating characteristic curve (AUC) obtained is 0.9 which implies the robustness of this technique. Primary contribution of the paper is personalized feature set for driver authentication. We provide performance comparison of several machine learning algorithms such as SVM, Random Forest, Naive Bayes, MLP etc. for driver authentication.