Gridwave: a grid-based clustering algorithm for market transaction data based on spatial-temporal density-waves and synchronization

The notion of density has been widely used in many spatial-temporal (ST) clustering methods. This paper proposes the novel notion of an ST density-wave, which is an extension of the notion of density. It also presents a new grid-based ST clustering algorithm called Gridwave based on the notion of ST density-waves and ST synchronization. The proposed algorithm can be used to discover synchronized changes in density among various locations as well as distinguish ST events and noise from market transaction data. Based on the theory of small-world networks, our algorithm can be used to evaluate ST synchronized correlations among regions with respective to the ST density over the whole network. To improve its performance, the proposed algorithm was implemented using parallel computing. To verify its feasibility, a real large-scale market transaction dataset was used to demonstrate the ST synchronized correlations and the final clustering results. Although our algorithm is applied in a domain-specific case, we suggest that the clustering notion and method could be generalized for other domain applications with similar ST data.