Colloquia Archive
A Bayesian Network Spatial Scan Statistic
April 22, 2008
AbstractMethods for spatial cluster detection attempt to locate spatial subregions of some larger region where the count of some occurrences is higher than expected. Event surveillance consists of monitoring a region in order to detect emerging patterns that are indicative of some event of interest. In spatial event surveillance, we search for emerging patterns in spatial subregions.
A well-known method for spatial cluster detection is Kulldorff's spatial scan statistic, which directly analyzes the counts of occurrences in the subregions. Neill et al. developed a Bayesian spatial scan statistic called BSS, which also directly analyzes the counts. We present a new Bayesian network spatial scan statistic called BNetScan, which models the causal relationships among the occurrences of interest and the observable occurrences using a Bayesian network. We evaluate the performance of the system on semi-synthetic data.
It is an open question whether we can obtain acceptable results using a Bayesian network if the probability distributions in the network do not closely reflect reality. We describe the results of experiments, which test the robustness of BNetScan relative to the probability distributions used to generate the data in the experiments. The experiments concern spatial event surveillance. As a point of comparison, we included Kulldorff's spatial scan statistic and BSS in the study. Our results indicate that BNetScan outperforms the other methods and that it is very robust relative to the probability distribution generating the data.
Xia Jiang, M.S., PhD candidate, School of Medicine, Department of Biomedical Informatics
Daniel B. Neill, Ph.D., Assistant Professor of Information Systems, Heinz School of Public Policy and Management
Carnegie Mellon University
Gregory F. Cooper, M.D., Ph.D., Associate Professor of Medicine and Intelligent Systems, School of Medicine, Department of Biomedical Informatics University of Pittsburgh
BiographyMs. Jiang is a Biomedical Informatics Ph.D. candidate in the Department of Biomedical Informatics, School of Medicine, University of Pittsburgh. Ms.Jiang has a master’s degree in computer science and a master’s degree in mechanical engineering. Before entering the Ph.D. program at the University of Pittsburgh, she taught for three years in the Department of Computer Science, Northeastern Illinois University. She co-authored “A Tutorial on Learning Causal Influences,” which appeared in the edited volume Innovations in Machine Learning, Springer-Verlag in 2006.She also co-authored the text Probabilistic Methods for Financial and Marketing Informatics, which was published by Morgan Kaufmann in early 2007.
While pursuing her Ph.D. at the University of Pittsburgh, she has concentrated on NSF funded research concerning the mining of clinical and over-the-counter data and the development of computational and statistical methods for the purpose of improving early detection and characterization of disease outbreaks. Results of her research have been presented at the Conference of the American Association for Artificial Intelligence in 2006, the Syndromic Surveillance Conference in 2006, and the Symposium of American Medical Informatics Association 2007. Some of the subsequent research work was described in papers submitted respectively to Journal of Biomedical Informatics, Journal of Statistic in Medicine, and International Journal of Approximate Reasoning. These papers are currently under review. Some of her work was included as book chapters in Applications of Bayesian Networks, Springer-Verlag, NY, 2008 and Scan Statistics - Methods and Applications, Birkhauser, 2008. Currently, she is co-authoring the text Probabilistic Bioinformatics Using Bayesian Networks. This text will cover applications of probability and statistics, in particular Bayesian networks, to bioinformatics.
