Using moment invariants to analyze cluster shapes and hypothesize potential causes

Although there are many algorithms and statistical tests to detect clustering of geographical phenomena, such as disease cases, the follow-up task of analyzing the cluster to explain its existence and mitigate the cluster is generally left to the researcher. These cluster detection methods are useful only for part of the process of identifying, understanding, and mitigating disease clusters. This research develops and presents a computer program that uses the information about the shape of the cluster to evaluate the hypotheses about potential causes. To achieve this, the shapes of the clusters are represented by image moment invariant statistics developed in the field of computer vision, and these shape statistics are compared against a database of moment invariants of shapes representative of several geographical processes, such as diffusion along roads and wind diffusion. Experiments using simulated data of different types of disease transmission were carried out, and the ability of the program to accurately classify the different types of diffusion demonstrates the viability of this approach to automated cluster analysis.