A top-down algorithm with free distance parameter for mining top-k flock patterns

Abstract

Spatiotemporal data is becoming more and more available due to the increase in the using of location-based systems. With such data, important information can be retrieved, where co-movement patterns stand out in finding groups of moving objects moving together. However, such pattern mining algorithms are not simple and commonly require non-trivial fixed parameters as input, which are extremely dependent on the data domain and also impacted by many others context variables, being such challenging task also to domain specialists. One example of these patterns is the flock pattern that has as its most challenging parameter the distance threshold that is the size of the disks that involves the objects. Although other density-based approaches reduce the impact of the restrictions of the disk, all of them still require a distance parameter for the density connectedness. Addressing this problem, we introduce the concept of discovering of k-co-movement patterns, which is finding the top-k patterns, according to the desired raking criterion. Especially for the flock pattern, we also define a new flock pattern query and propose a top-down algorithm with free distance parameter for the aforementioned problem.