Performance Evaluation of Co-location Miner

Given a collection of boolean spatial features, the co-location pattern discovery process nds the subsets of features frequently located together. For example, the analysis of an ecology dataset may reveal the frequent co-location of a re ignition source feature with a needle vegetation type feature and a drought feature. The spatial co-location rule problem is diierent from the association rule problem. Even though boolean spatial feature types (also called spatial events) may correspond to items in association rules over market-basket datasets, there is no natural notion of transactions. This creates diiculty in using traditional measures (e.g. support, conndence) as well as association rule mining algorithms using support based pruning. We recently deened the problem of mining spatial co-location patterns and proposed the Co-location Miner 25], an algorithm for mining co-locations. In this paper, we present an experimental performance evaluation of Co-location Miner. For the purpose of comparison, we consider two other approaches, namely the pure geometric approach and the pure combinatorial approach. Empirical evaluation shows that the pure geometric method performs much better than the pure combinatorial method when generating size 2 co-locations; however, it becomes much slower when generating co-locations with more than 2 features. Co-location Miner integrates the best features of the above two approaches and provides the best overall performance. Experimental results also show that Co-location Miner is robust in the face of noise and scales up gracefully with increases in the number of spatial feature types, maximum size of co-location patterns, and the number of instances of spatial features.