Forecasting earthquakes by Machine Learning techniques: lights and shadows

Abstract Earthquake prediction research continues to be a very active research topic because its correct prediction could save many human lives as well as the impact it has on the economy of a country or region. Furthermore, the prediction of earthquakes, both in terms of location, time, magnitude and probability also continues to be a topic of interest in research due to the difficulty of predicting them, precisely because of the high non-linearity of their behaviour.This is why Machine Learning methods are becoming increasingly popular for the challenging task of earthquake prediction. The methods presented limit their application to one geographical area, which makes sense because of the different structure of the earth's crust that greatly affects earthquakes, and with spectacular results. However, could these methods be applied to other geographical locations with good results? The results obtained, if good, could they be applied to the real behaviour of seismic phenomena? This work answers these questions and compares its results with those found in the literature, and concludes that the method chosen for the comparison does not present a correct behaviour in a real environment.