Machine Learning-based Intelligent Prediction Of Mechanical Properties Of The Strata Rocks At Thar Coalfield In Pakistan

Mechanical properties of rocks,such as uniaxial compressive strength（UCS）in MPa and Elastic modulus（E）in GPa are among the important data in mining engineering planning and design and play a critical role in the design of mining rock mechanic projects.UCS and E of rock have broad applications in the stability of surface and underground structures in mining.Intelligent indirect techniques for designing and excavating underground structures make use of a limited amount of data for design,saving time and money while ensuring the stability of the mining structures.This study has economic and even social implications,which are integral elements of sustainability.Moreover,this thesis aims to estimate accurately crucial rock mechanical parameters of the overlying and surrounding rock strata,especially UCS and E which are necessary in order to better understand the underlying mechanisms for designinig mining structures and earth surface profile that might otherwise adversely affect the mining environment,safe coal production,mining sustainability and so on.UCS and E can be determined directly by conducting laboratory tests,which are time consuming,and require high-quality core samples and costly modern instruments.Thus,devising an indirect estimation method of UCS and E has promising prospects.In this study,eight robust machine learning（ML）-based intelligent prediction models,namely,light gradient boosting machine（Light GBM）,support vector machine（SVM）,category boosting（Catboost）,artificial neural network（ANN）,adaptive neuro-fuzzy inference system（ANFIS）,gradient boosted tree regressor（GBRT）,random forest（RF）,and extreme gradient boosting（XGBoost）,were developed to predict the impacts of four input parameters,namely,wet density(ρ_wet)in g/cm³,moistureω（%）,dry density（ρ_d）in g/cm³,and Brazilian tensile strength（BTS）in MPa on output UCS and E.The associated strengths of every input and output were systematically measured employing a series of fundamental statistical investigation tools to categorize the most dominant and important input parameters.The original dataset of 106 samples of UCS and E was split as 70%for the training and 30%for the testing for each model.In order to enhance the performance of each developed model,an iterative 10-fold cross-validation method for UCS and a 5-fold cross-validation method for E was used.Therefore,based on the results of the study,the XGBoost model outperformed the other developed models with higher accuracy,coefficient of determination（R²=0.999）,mean absolute error（MAE=0.00054）,mean square error（MSE=0.000005）,root mean square error（RMSE=0.00069）,and a20 index=0.999 for UCS and R²=0.999,MAE=0.0015,MSE=0.0008,RMSE=0.0089,and a20 index=0.996 for E,respectively,on the test data.In addition,GBRT and RF have also shown high accuracy in predicting UCS and E,with R² values of0.941 and 0.973 and 0.988 and 0.989,respectively,on the test data,but they can be used conditionally.Based on sensitivity analysis,all parameters were positively correlated,while BTS was the most influential parameter in predicting UCS and E.Using an ML-based intelligent approach,this study could be able to provide alternative elucidations for predicting UCS and E with appropriate accuracy and run time at Thar coalfield of Pakistan.