TRACKING THE MOVING PATH OF CLUSTER WITH ABNORMAL INCIDENCE RATE

• Main Authors: Chu-Yi Lo, 羅筑儀
• Other Authors: Chen-Ju Lin
• Format: Others
• Language: zh-TW
• Online Access: http://ndltd.ncl.edu.tw/handle/27786281304785074086

碩士 === 元智大學 === 工業工程與管理學系 === 103 === This research aims at detecting the abnormal region with increased incidence rate. The number of incidence is considered as the spatio-temporal data following Poisson distribution. The areas that simultaneously have increased incidence rate in a neighborhood region are regarded as an abnormal cluster. The goals of this research are to detect the location, coverage size, and to track the moving path of an abnormal cluster. Scan statistic is a popular method for analyzing the spatio-temporal data. The method is used for detecting the existence of an abnormal cluster. Although the scan statistic is able to signal sustained anomalies, continually applying scan statistic cannot monitor the moving direction of an abnormal cluster. Scan Statistic can only signal the existence of clustering without specifying its exact location. Thus, such method cannot be continuously applied to monitor the moving path of abnormal clusters. Hence, this research applies the Kalman Filter (KF) that commonly used in the image and signal tracking domains to track the moving path of the cluster with abnormal incidence rate. This research presents a tracking scheme Maximum Likelihood Kalman Filter (MLKF). The research method firstly uses the spatial scan statistic to determine the existence of an abnormal cluster with increased incidence rate. Then KF is applied to calculate the most likely position of the cluster, and to predict its moving path. The research simulated several moving patterns with different moving velocity, moving direction, cluster size and cluster location. Under the simulated moving patterns, continuously applying spatial scan statistic only has 30% of chance to signal clustering after the first signal. However, MLKF can continuously locate the abnormal cluster with satisfying accuracy. The accuracy of MLKF mainly depends on cluster coverage size and the initial parameter settings of MLKF, but is less effected by moving velocity and direction of a cluster.