1. Poisson hidden markov model on earthquake occurrences in Metro Manila, Philippines.
- Author
-
Felix, Edd Francis O., Buhat, Christian Alvin H., and Mamplata, Jonathan B.
- Subjects
- *
HIDDEN Markov models , *EARTHQUAKES , *BOX-Jenkins forecasting , *TYPHOONS , *EXPECTATION-maximization algorithms , *STOCHASTIC models , *EARTHQUAKE zones - Abstract
The Philippines, as part of the Circum-Pacific belt, is considered as one of the most seismically active countries in the world. Earthquake occurrence is frequent and its effects vary depending on its size. Understanding how the occurrences happen is therefore important. Stochastic models of earthquake occurrence have been used to study seismic activities in various active earthquake zones globally. In this paper, we apply Poisson hidden Markov models (PHMM) using the January 1, 1960 to January 20, 2019 earthquake data of Metro Manila, Philippines. The parameters in the models are estimated using expectation-maximization (EM) algorithm. We determine using various statistical tests that the 5-state PHMM best represents the earthquake data and implement bootstrap algorithm to validate the acceptability of its parameter estimates. Moreover, we investigate the forecasting capability of the 5-state PHMM by comparing it to the ARIMA model. Using unscaled mean bounded relative absolute error (UMBRAE), we find that the 5-state PHMM gives closer one-step ahead forecasts and is a better forecasting model for the considered data. [ABSTRACT FROM AUTHOR]
- Published
- 2022
- Full Text
- View/download PDF