Item

Abstract

The advances in machine learning have led to a renewed interest in the challenging task of earthquake prediction. Analysis of the global geomagnetic field has shown promising results in earthquake precursors study. However, applying this approach to develop an accurate earthquake prediction model has proven to be difficult partly because of the sophistication of the data. This study presents an Automatic Machine Learning (AutoML) technique for handling complex geomagnetic data and imbalance earthquake events by selecting the optimum machine learning model through Bayesian optimization. Global geomagnetic data and earthquake events that occurred between 1970 to 2020 was used in this study. Sampling techniques approaches showed promising results to solve the problem of an imbalance dataset, essential features were extracted and a multi-class classification model was developed using sample dataset that consists of oversampling and undersampling data. Sample and Raw models were developed from two different dataset, the result shows that the Sample model selects Neural Network as the best model, outperforms Raw model. Raw model shows 94% accuracy, 44% Matthew Correlation Coefficient (MCC) and 0% F-1 score, meanwhile Sample model shows 89% accuracy, 86% MCC and 89% F1-score. It is found that a neural network multi-class classification model addresses the solution to earthquake prediction challenges when using global geomagnetic data.