A novel similarity measure for multiple aspect trajectory clustering

Multiple aspect trajectories (MATs) is an emerging concept in the domain of Geographical Information Systems, where the basic view of semantic trajectories is enhanced with the notion of multiple heterogeneous aspects, characterizing different semantic dimensions related to the pure movement data. Many applications benefit from the analysis of multiple aspects trajectories, ranging from the analysis of people trajectories and the extraction of daily habits to the monitoring of vessel trajectories and the detection of outlying behaviors. This work proposes a novel MAT similarity measure as the core component in a hierarchical clustering algorithm. Despite the many clustering methods in the literature and the recent works on MAT similarity, there are still no works that dig deeper into the MAT clustering task. The current article copes with this issue by introducing TraFoS, a new similarity measure that defines a novel method for comparing MATs. TraFos includes a multi-vector representation of MATs that improves their similarity comparison. TraFos allows us to compare MATs across each aspect and then combine similarities in a single measure. We compared TraFos with other state of the art similarity metrics in Agglomerative clustering. The experimental results show that TraFos outperforms other similarities metrics in terms of internal, external clustering metrics and training time.