Code & Dataset for 'Predicting Stick-Slips in Sheared Granular Faults Using Machine Learning Optimized Dense Fault Dynamics Data'

Predicting Stick-Slips in Sheared Granular Faults Using Machine Learning Optimized Dense Fault Dynamics Data Weihan Huang, Ke Gao, Yu Feng Key Points · LightGBM model is trained using densely distributed fault dynamics data to predict laboratory earthquakes in a sheared granular fault system · The dense fault dynamics data contain sufficient information for prediction; the prediction accuracy can be improved via data optimization · The SHAP value analysis of input data has the potential to unveil the complex underlying fault physics in sheared granular fault The laboratory data simulation used in this study are publicly available in this site. The FDEM simulation datasets are saved as file 'p28_data.bin'. The python program of machine learning algorithm to train and test the data is saved as 'submission_lgbm_model.py'. Besides, the basic parameters setting and functions using in the program are saved as 'submission_setting.py' and 'submission_function_define.py'.