Clustering driving trip trajectory data based on pattern discovery techniques

Identifying patterns to characterize driving human driving styles from driving trip data is a promising and interesting area of research and application. To cluster the driving trips in a set of recorded GPS tracks, this paper presents an information theoretic approach to characterize them based on their occurrences of frequently detected patterns. The patterns are discovered through a statistical significance test on a generated set of spatio-temporal data and its associated attributes that represent the characteristics of recorded GPS data. For evaluating the performance of the proposed approach, a real dataset with ground truth information is tested to validate its clustering power and compare with other approaches. The result indicates the approach is effective and efficient to extract interpretable features to summarize the complex driving behaviors to form a good representation of driving styles for machine learning to achieve good performance.