UltraMovelets: Efficient Movelet Extraction for Multiple Aspect Trajectory Classification

Due to the fast-paced advances in mobile devices and the Internet, mobility data volume has grown significantly in the last few years. The explosion of social media data, sensors, IoT, and Internet-enabled sources allowed the semantic enrichment of such mobility data, which evolved from raw spatio-temporal data to high dimensional semantic data. This new scenario makes mobility analytics tasks, such as classification, more challenging due to the diverse and intricate nature of the data. Classification methods capable of dealing with multiple semantic aspects are based on the concept of movelets, the parts of a trajectory that better discriminate a class and can improve classification accuracy. However, existing movelet extraction methods could be more computationally efficient, making them more feasible in high-dimensional datasets. In this work, we propose UltraMovelets, a new algorithm for movelets extraction that is more efficient than the existing methods while improving or maintaining classification performance. We evaluated our method using the accuracy metric to evaluate the classification performance and scalability to measure processing time and memory usage. Experimental results obtained from well-known trajectory-user-linking datasets indicate that UltraMovelets exhibit an improvement in accuracy at the expense of running time. Notably, the observed improvements vary across datasets, emphasizing the nuanced nature of performance enhancements. The results indicate that UltraMovelets outperformed MASTERMovelets in terms of speed, and, in scalability experiments, it showed better performance compared to HiperPivots. By reducing the search space, UltraMovelets use fewer computational resources.