Driver classification with using machine learning methods on vehicle driving data

The advances in vehicle equipment technology enabled us collecting large-scale vehicle driving data. This data is an important resource for urban area traffic management and vehicle driving support system applications. In this study, we are interested making inferences ability of these data about the driver. Uyanık data set [1] CAN (Controller Area Network) bus data are used as the data source. Experiments are carried out on driver gender classification, driver identification, and driver clustering. In the driver clustering experiments, hierarchical driver clustering is performed with using the Dynamic Distance Warping developed by us and Dynamic Distance Time data conversion methods according to different distance metrics. As a result, consistent driver groupings are achieved. In driver gender classification experiments, gender classification is performed with applying data sampling, feature extraction, feature elimination and discretization data preprocessing methods. Best classifiers reached up to 0.97 accuracy rate. In driver identification experiments, driver classification is carried out with applying driver pattern splitting technique developed by us and feature extraction data preprocessing methods and driver identification performance reached 0.1 accuracy rate among the 105 drivers. All these experiment results open up a new thread of discussion: whether the driving data should be treated as a sensitive personal feature?