Deep learning algorithms to forecast the magnitude of earthquakes in Peru

Abstract

The research proposes the application of the LSTM algorithm with the objective of forecasting the magnitude of the earthquakes in Peru by adding pink noise in their dynamics to replicate and capture the complex dynamics of the magnitude, an antipersistence in the increases is confirmed by means of the coefficient of hurst. less than 0.5 and stable. An LSTM network model of higher accuracy in generating forecasts than the single-direction multilayer perceptron network is then implemented. Therefore, evidence of a better performance of the LSTM network is presented in relation to the earthquake magnitude variables due to the pink noise that occurs in its dynamics. The study period runs from January 1, 2020 to October 23, 2021.