USID - Unsupervised Identification of the Driver for Vehicle Comfort Functions

High comfort is one of the main demand for a modern vehicle. Comfort functions that are designed to ensure the comfort of modern vehicles are becoming more tailored to each driver. In order to maximize the effectiveness of comfort functions the driver must be precisely identified. The driver identification task can be accomplished by utilizing vehicle data from the standardized On-Board Diagnostic II system (OBD II).
In this paper, the feasibility of precise driver identification was investigated based on unsupervised machine learning methods. The authors propose the USID (Unsu-pervised Identification of the Driver) concept for this purpose. The USID promises rich scalability since the unsupervised models don’t use predefined classes to identify drivers. The unsupervised methods used in this work are K-Means, Autoencoders, Self-Organizing maps, and Density-based spatial clustering (DBSCAN). The models are trained and evaluated using the OBD II time series of 16 drivers driving the same vehicle. In the end, the experimental analysis of the USID was done that showed very good confidence of the concept in driver identification during the driving cycle of all 16 drivers.