Detection of irregular-shaped clusters on a network by controlling the shape compactness with a penalty function

Recent development of cluster detection methods focuses on the improvement of efficiency or accuracy, with the latter yielding a wide range of variants in the shape of the search window, from a simple circle and elliptic shape to more irregular shapes. Detection of irregular-shaped clusters has seen various new approaches as it is considered to capture the shape and extent of clusters more accurately. One of these newly developed approaches achieves the irregularity of the clusters by placing a penalty on the shape complexity of a candidate cluster. This study extends this approach and applies it to a network-space to detect irregular-shaped clusters along a street network segments in a small urban area. The study uses a genetic algorithm to search candidate clusters and identify the most likely cluster using the framework of spatial scan-statistics. Application of the method to a small synthetic data and a real data set revealed that providing options of different cluster patterns with different compactness parameters helps find more accurate as well as geometrically and contextually more meaningful clusters, as opposed to those detected without a shape controlling parameter.