Parallel approach to incremental co-location pattern mining

Abstract

Co-location pattern mining is an important area of spatial data mining. In many real word applications, new data is continuously arriving to the system and is stored in spatial databases. As co-location discovery is computationally demanding task, it is crucial to maintain co-location patterns for such dynamic databases without recalculating them from scratch. In this paper we present the first GPU-parallelized solution to this problem. Our contribution is threefold: 1) we present a modified version of EUCOLOC algorithm using the iCPI-tree method called iCPI-EUCOLOC, 2) we modify state-of-the-art MGPUCPM algorithm to implement iCPI-EUCOLOC algorithm on GPUs and 3) we present experimental results showing large performance improvements over the original MGPUCPM algorithm. Our solution allows to reduce user waiting times and is economically beneficial due to the reduced overall computation time.