A1233
Title: Statistical detection of arbitrarily shaped spatial clusters in continuous data
Authors: Fumio Ishioka - Okayama University (Japan) [presenting]
Abstract: Detecting spatial clusters is a critical task in many scientific fields. Conventional spatial scan statistics, while powerful, are primarily designed for count data and are limited to simple geometric shapes. A statistical method is proposed for identifying arbitrarily shaped clusters in continuous spatial data. The approach integrates the spatial scan statistic framework with echelon analysis to systematically generate a hierarchy of candidate clusters, from which the most significant one is identified using a likelihood ratio test. Effectiveness and robustness are evaluated through extensive numerical simulations under diverse conditions, including varying mean differences, overall data uncertainty, and intracluster uncertainty. Performance is also examined for different cluster shapes, including circular, linear, and ring-shaped patterns. Additionally, robustness is investigated using data generated from several non-normal distributions, including Laplace, Poisson, and lognormal distributions. Overall, the proposed framework offers a flexible and practical approach for detecting spatial clusters in continuous spatial data.
