FTS: a feature-preserving trajectory synthesis model

Abstract

Driven by the GPS-enabled devices and wireless communication technologies, the researches and applications on spatio-temporal databases have received significant attentions during the past decade. Hence, large trajectory datasets are extremely necessary to test high performance algorithms for these applications and researches. However, real-world datasets are not accessible in many cases due to privacy concerns and business competition. For this reason, we propose a feature-preserving model FTS to generate new trajectories in this work. The proposed model is composed of three components: 1) Extracting data features from the original dataset. 2) Generating new trajectories. 3) Validating the result by comparing the features of generated trajectories with the given dataset. However, it is hard to make the diverse features of generated dataset consistent with those of original dataset. To tackle this challenging problem, we present several novel algorithms in this paper. Extensive experiments based on real trajectory datasets exhibit that the synthetic datasets generated by FTS preserve the features of original datasets successfully.