Fuzzy feature cluster-based approaches for discovering spatial high-utility co-location patterns

Co-location patterns describe sets of geographically distributed features that maintain consistent spatial proximity in neighboring regions. As an important extension in spatial co-location pattern mining, spatial high-utility co-location patterns move beyond traditional frequency-based metrics to ensure high utility of mining outcomes. Most existing methods employ the utility participation index (UPI) to evaluate pattern utility. However, UPI neglects the fuzziness of neighbor relationships in real geographic environments, and does not distinguish the degree of proximity between instances. It also overlooks issues such as distance decay and instance sharing, while incurring computational overhead due to violation of the Apriori principle. To address these limitations, we introduce binary fuzzy set theory to model proximity relationships, using fuzzy membership degrees to represent instance proximity. For utility evaluation, we incorporate both distance attenuation and instance sharing, proposing a novel measure for feature-pair utility. Furthermore, we develop two clustering algorithms to extract high-utility co-locations from fuzzy feature clusters. Experimental results validate the theoretical soundness and computational efficiency of our approach.