Mining Trajectory Databases via a Suite of Distance Operators

Abstract

With the rapid progress of mobile devices and positioning technologies, Trajectory databases (TD) have been in the core of database research during the last decade. Analysis and knowledge discovery in TD is an emerging field which has recently gained great interest. Extracting knowledge from TD using certain types of mining techniques, such as clustering and classification, impose that there is a mean to quantify the distance between two trajectories. Having as a main objective the support of effective similarity query processing, existing approaches utilize generic distance metrics that ignore the peculiarities of the trajectories as complex spatio-temporal data types. In this paper, we define a novel set of trajectory distance operators based on primitive (space and time) as well as derived parameters of trajectories (speed and direction). Aiming at providing a powerful toolkit for analysts who require producing distance matrices with different semantics as input to mining tasks, we develop algorithms for each of the proposed operators. The efficiency of our approach is evaluated through an experimental study on classification and clustering tasks using synthetic and real trajectory datasets.