Magnitude Prediction Model for Japan Seismic Tremors Using Artificial Neural Network

Abstract

The artificial neural network (ANN) model for predicting seismic tremor magnitudes for Japan is portrayed in this paper. The authors have retrieved the earthquake dataset from European-Mediterranean Seismological Center for this study. The dataset comprises a list of 5000 quake events that occurred from 1st July 2010 to 14th April 2016 in the region of Japan. Different neural network structures and ANN configurations exemplify the ANN model construction. The experiment is carried out by fine-tuning network variables such as type, transfer function, training function, and hidden neurons. The forecast accuracies of each of these network configurations are compared. The resultant ANN model features Levenberg-Marquardt backpropagation method for training the model, the nonlinear sigmoid activation function for the hidden layer, and the model's performance is evaluated concerning Mean Squared Error (MSE) and Gradient (g).