Machine learning for prediction of the uniaxial compressive strength within carbonate rocks

Abstract

The Uniaxial Compressive Strength (UCS) is an essential parameter in various fields (e.g., civil engineering, geotechnical engineering, mechanical engineering, and material sciences). Indeed, the determination of UCS in carbonate rocks allows evaluation of its economic value. The relationship between UCS and numerous physical and mechanical parameters has been extensively investigated. However, these models lack accuracy, where as regional and small samples negatively impact these models' reliability. The novelty of this work is the use of state-of-the-art machine learning techniques to predict the Uniaxial Compressive Strength (UCS) of carbonate rocks using data collected from scientific studies conducted in 16 countries. The data reflect the rock properties including Ultrasonic Pulse Velocity, density and effective porosity. Machine learning models including Random Forest, Multi Layer Perceptron, Support Vector Regressor and Extreme Gradient Boosting (XGBoost) are trained and evaluated in terms of prediction performance. Furthermore, hyperparameter optimization is conducted to ensure maximum prediction performance. The results showed that XGBoost performed the best, with the lowest Mean Absolute Error (ranging from 17.22 to 18.79), the lowest Root Mean Square Error (ranging from 438.95 to 590.46), and coefficients of determination (R2) ranging from 0.91 to 0.94. The aim of this study was to improve the accuracy and reliability of models for predicting the UCS of carbonate rocks.