LAUSR

Detecting irregularly shaped spatial clusters within heterogeneous point processes is challenging because the number of potential clusters with different sizes and shapes can be enormous. This research develops a novel method, expansion-based spatial clustering for inhomogeneous point processes (ESCIP), for detecting spatial clusters of any shape within a heterogeneous point process in the context of analyzing spatial big data. Statistical testing is used to find core points—points with neighboring areas that have significantly more cases than the expectation—and an expansion approach is developed to find irregularly shaped clusters by connecting nearby core points. Instead of employing a brute-force search for all potential clusters, as done in the spatial scan statistics, this approach only requires testing a small neighboring area for each potential core point. Moreover, spatial indexing is leveraged to speed up the search for nearby points and the expansion of clusters. The proposed method is implemented with Poisson and Bernoulli models and evaluated for large spatial data sets. Experimental results show that ESCIP can detect irregularly shaped spatial clusters from millions of points with high efficiency. It is also demonstrated that the method outperforms the spatial scan statistics on the flexibility of cluster shapes and computational performance. Furthermore, ESCIP ensures that every subset of a detected cluster is statistically significant and contiguous. Key Words: cyberGIS, spatial algorithm, spatial analysis, spatial clustering.