PROJECT TITLE :

Stat-DSM: Statistically Discriminative Sub-Trajectory Mining With Multiple Testing Correction

ABSTRACT:

We propose a novel statistical approach, which we call Statistically Discriminative Sub-trajectory Mining (Stat-DSM), to evaluate the statistical significance of the results from discriminative sub-trajectory mining. Another name for this method is statistical reliability. The purpose of the Stat-DSM algorithm, which is given two groups of trajectories, is to identify moving patterns, in the form of sub-trajectories, that are observed statistically significantly more frequently in one of the groups than in the other. The statistical significance of the extracted sub-trajectories is properly controlled thanks to the proposed method. This means that the probability of finding a falsely discriminative sub-trajectory is lower than a specified significance threshold (for example, 0.05), which is essential when the method is used in scientific or social science studies that take place in noisy environments. The task of finding statistically discriminative sub-trajectories within a massive trajectory dataset presents challenges both from a computational and a statistical point of view. We address these challenges in the Stat-DSM method by presenting a tree representation of the sub-trajectories, and then applying an effective statistical inference method that is based on permutations to the tree. According to our best knowledge, Stat-DSM is the first method that offers a statistical approach to quantify the reliability of the results of discriminative sub-trajectory mining. By applying the Stat-DSM method to a real-world dataset that contains one million trajectories, we are able to demonstrate both the efficacy of the method as well as its scalability.