Using Hidden Markov Models for Profiling Driver Behavior Patterns

Detection of risky driving events using smartphone based sensing is a growing technology devoted to impact positively driving behaviors. This technology might improve traffic and reduce the number of car accidents. However, data measured from inbuilt smartphone sensors, represented as a multivariate time series, commonly contains strong temporal dynamics. As a result, there is a growing need for developing methods able to handle such dynamics to make any inference based on the data in hand. In this work, we present a methodology for discriminating risky driving events from smartphone sensors using Hidden Markov Models, which are a well known statistical method for dealing with time-series comprising time-varying information. The methodology is validated using a publicly available dataset, where we demonstrated that the achieved results are comparable with state-of-the-art approaches, yielding accuracy rates around 90% in a seven classes problem.