Multi-Target Regression Using Convolutional Neural Network-Random Forests (CNN-RF) For Early Earthquake Warning System

Abstract

Indonesia occupies a very active tectonic zone because the world's three large plates and nine other smaller plates meet each other in Indonesian territory and form a complex plate meeting path. East Java Province is part of the Sundanese arc, which has a relatively high level of seismicity and has a complex geological system resulting from the Indo-Australian plate. Therefore, a system that can provide earthquake early warning (EEW) is needed to reduce casualties. In this paper, we determine the epicenter and magnitude of the earthquake using the Multi-Target Regression Convolutional Neural Network-Random Forest (CNN-RF). This model uses Multi-Target Regression Convolutional Neural Network (CNN) as feature extraction and Multi-Target Regression Random Forest (RF) for multi-target regression. Earthquakes in East Java in 2009-2017 are used to train and validate the proposed model. Based on the experiment, the lowest error obtained from the Multi-Target Regression CNN-RF model is 16.3 km for longitude, 36.4 km for latitude, and 0.3095 for magnitude.