Auto-REP: An Automated Regression Pipeline Approach for High-efficiency Earthquake Prediction Using LANL Data

Abstract

Earthquake prediction, which is a key issue that has long existed among seismologists, is of high scientific importance. An earthquake prediction model can output the time of earthquake occurrence in advance using machine learning methods, which is receiving increasing attention. Earthquake prediction involves a large variety of data mining steps, which requires a large amount of time for processing data and model development. Thus, an efficient and accurate prediction method is needed. Aiming to solve this problem, we propose Auto-REP, an automated machine learning-based regression model. Our contribution of Auto-REP is using laboratory seismic data to develop a regression pipeline in an automated manner, and eventually obtain the prediction results of laboratory earthquake occurrence. The automated pipeline consists of feature extraction, feature selection, modelling algorithm and optimization. With this approach we extract features from each of the earthquake channels which results in a massive feature space. The hyperparameters of the model are optimized by a Bayesian technique as part of the automated approach. The experimental results shows that the MAE and MSE of our model in the training and testing datasets are 1.48, 1.51 and 1.52, 1.59. The results demonstrate that our Auto-REP method can predict laboratory earthquakes efficiently and accurately.