Markov modulated poisson process for anomaly normalization scheme in public complaint system

Abstract

The Government of Jakarta uses an online — mobile based systems which allow fast and transparent response of citizen's complaints. However, an initial investigation indicates anomalous activities ranging from bursty to repetitive events. These anomalies could potentially waste government agencies and public servant resources. Furthermore, quality level of analysis which produced from ”raw” data will raise a lot of questions. A Markov Modulated Poisson Process (MMPP) are proposed for anomaly detection and normalization scheme. The model consists of a time-varying Poisson process that includes seasonal variation in Poisson rates over time, as well as a Hidden Markov event process. MMPP able to model posterior probability of anomalous non-homogeneous Poisson events as function of time for assessing normal and unusual bursty events. Moreover, posterior probability value enable normalization process to be performed. Simulation results shows that proposed schemes performs better compared to baseline Poisson threshold test approach in terms of anomaly detection accuracy, precision, and recall performance.