Multimodal Behavioral Mobility Pattern Mining and Analysis Using Topic Modeling on GPS Data

摘要

Identifying risky driving behavior is of central importance for increasing traffic safety. This paper tackles the task of analyzing real (naturalistic) driving data captured by in-vehicle sensors using interpretable data science methods. In particular, we focus on symbolic time-series abstraction and the subsequent behavioral profile identification using topic modeling approaches. For our experiments, we utilize a real-world dataset. Our results indicate interesting behavioral driving profiles including important patterns and factors for traffic safety modeling.