Prediction of Earthquake Magnitude Using Tree-Based Ensemble Learning

The present paper aims at predicting earthquake magnitude (Mw) from soil gas radon concentration (CRn) and three meteorological parameters (hourly humidity (H), hourly temperature (T), and hourly air pressure (P)). To accomplish this, three tree-based ensemble machine learning approaches, namely gradient boosting (GBoost), extreme gradient boosting (XGBoost), and random forest (RF), were created. A total of 386 datasets including recorded Mw values and measured soil gas CRn, H, T, and P values were used to develop the models. The models were then verified using statistics such as relative absolute error (RAE), root mean square error (RMSE), mean absolute error (MAE), and the ratio of RMSE to data standard deviation (RSR). A comparison of the performance metrics reveals that the GBoost model predicted the Mw value with lower MAE, RMSE, RSR, and RAE values than both the XGBoost and RF models. Performance was also verified using rank analysis and plots of Taylor and scaled percentage error (SPE). Rank analysis showed that the GBoost model received higher overall scores than both the XGBoost and RF models, indicating that the GBoost model achieved better prediction accuracy than both the XGBoost and RF models in predicting the Mw values. Both Taylor and SPE plots showed that GBoost model predicted the Mw values more accurately than XGBoost and RF models. According to the results of the study, the GBoost model can be used to predict Mw reliably and quickly, provided that three meteorological factors (H, T, and P) and soil gas CRn values are available.