Earthquake forecasting in the Himalayan region using neural networks models

Earthquake forecasting using neural networks models is presented in the study. The problem of earthquake forecasting was modelled as a pattern classification task, with two output class labels representing the monthly occurrence and the non-occurrence of the event. The input features are the parameters of the Gutenberg-Richter relation and Characteristic earthquake model, which were computed using the historical earthquake data available in the form of earthquake catalogues. A total of nine features representing the seismic activity were selected. The Himalayan region divided into seven source zones is the area considered for the study. Six neural network models, Neural network with back-propagation algorithm, recurrent neural network, radial basis function neural networks, Levenberg–Marquardt algorithm, pattern recognition neural network, and logistic regression classifier were used for forecasting. The accuracy of the models was measured using the Hanssen-Kuiper skill score. The smaller magnitude earthquakes, magnitude between 5 to 5.5, were forecast more accurately over the large magnitude events by all the models.