Abstract
Current outbreak detection algorithms monitoring single data stream may be prone to false alarms due to baseline shifts that could be caused by large local events such as festivals or super bowl games. In this paper, we propose a Multinomial-Generalized-Dirichlet (MGD) model to improve a previously developed spatial clustering algorithm, MRSC, by modeling baseline shifts. Our study results show that MGD had better ROC and AMOC curves when baseline shifts were introduced. We conclude that MGD can be added to outbreak detection systems to reduce false alarms due to baseline shifts.