A1374
Title: Statistical inference for a contaminated zero-inflated compound Poisson-gamma process with applications
Authors: Shih-Feng Huang - National Central University (Taiwan) [presenting]
Abstract: The estimation problem for a contaminated zero-inflated compound Poisson-gamma (ZiCPG) process is considered. The contaminated ZiCPG model provides a flexible framework for modeling the dynamics of stochastic processes with jumps by simultaneously incorporating zero inflation and random jump sizes into a Poisson process with noise contamination. Consequently, computing the maximum likelihood estimator is challenging. To address this issue, an effective procedure based on jump-state identification and particle filtering is proposed to estimate latent jumps and mitigate the computational challenges arising from noise contamination. A prediction model is further developed based on the estimated jumps. Numerical results from various simulation scenarios demonstrate that the proposed estimation scheme achieves satisfactory performance. Moreover, when applied to an electricity consumption dataset, the proposed prediction model yields superior performance in forecasting potential abnormal electricity consumption compared with competing methods.
