Detection of Space-Time Cluster

Abstract

Detection of space-time cluster is an important aspect of spatial epidemiology and GIS-based data mining. This paper compares three clustering algorithm namely, scan statistic [1], local indicators of spatial autocorrelation (LISA) [2] and local G-statistic [3]. This study involves application of routine clinical service data collected by a Northeast Ohio healthcare organization in USA over a period 1994 -2006 to find excess space-time variations of lung cancer. Using empirical Byes adjustment of incidence rate, almost identical spatial pattern of clusters were detected by the three algorithms. However, the space-time scan statistics involving cylindrical search window shows somewhat different spatial localization. Finally, the study compares the effectiveness the different methods.