Optimizing Machine Learning Models for Predicting Rock Cohesion and Angle of Internal Friction: A Comparative Study of Lithological Analysis, Robustness Assessment, and SHAP Explanations

Applications of Machine Learning Models to Predict the Uniaxial Compressive Strength of Sandstone from Muzaffarabad - A Case Study

Real-Time Prediction of Rock Young Modulus and Uniaxial Compressive Strength from Artificial Intelligence-Based Correlations

This study investigates an artificial intelligence (AI) based approach for real-time prediction of Young's modulus and UCS using drilling and logging data and reveals that neutron porosity, formation bulk density and Gamma Ray are the three most influential predictors.

Estimation of Rock Shear Strength Parameters from Geophysical Indicators Using an Equilibrium-Optimized Multilayer Perceptron

Accurate and scalable estimation of rock shear strength parameters is essential for remote-sensing-supported geological hazard assessment, slope stability evaluation, and engineering geological mapping. However, determining cohesion and internal friction angle requires multiple triaxial tests under different confining pressures, which are time-consuming, costly, and difficult to apply widely. To support remote-sensing-oriented geoscience and civil engineering applications, this study develops a hybrid machine learning framework for estimating cohesion and internal friction angle from geophysical and mechanical indicators. A cross-source database was compiled from published rock records collected from the Jinchuan mining area in China and the Luhri area in India. After completeness screening and unit harmonization, 213 mixed-lithology cases were retained for modeling, with P-wave velocity, density, uniaxial compressive strength, and tensile strength used as input variables. An Equilibrium Optimizer was coupled with a multilayer perceptron to optimize the network weights and biases, and model performance was evaluated using five-fold cross-validation, independent testing, repeated runs, and comparisons with conventional MLP and several typical machine learning models. The proposed EO–MLP model achieved high prediction accuracy, with test-set coefficient of determination values of 0.946 for internal friction angle and 0.983 for cohesion and corresponding RMSE values of 1.072 and 0.671, respectively. Robust scaler normalization produced the best performance among the three tested normalization strategies. SHapley Additive exPlanations analysis indicated that density was the dominant predictor of cohesion, whereas uniaxial compressive strength and P-wave velocity made the largest contributions to internal friction angle prediction. The proposed framework provides an indirect data-driven tool for estimating shear strength parameters and can complement engineering-geological investigation and rock engineering design.

Comparative Evaluation of Intelligent Machine Learning Models for Prediction of Uniaxial Compressive Strength of Rocks

The results showed that all models successfully captured the relationship between the input parameters and uniaxial compressive strength, although their predictive capabilities differed considerably, and KStar produced the most accurate predictions.

The Evolution and Future of Machine Learning in Rock Slope Stability: A Systematic Review and Meta-Analysis

A Novel Rate of Penetration Prediction Model Integrating Log-Derived Geomechanical Properties and Physically Motivated Energy Parameters for Heterogeneous Formations

Accurate rate of penetration (ROP) prediction in heterogeneous formations remains a key challenge for drilling optimization, as existing empirical models rely on fixed-structure coefficients unable to adapt to rapid lithological transitions. This study presents a novel exponential-form ROP model integrating surface drilling parameters (weight on bit (W), rotary speed (N), torque (T), and standpipe pressure (SPP)), log-derived geomechanical properties (dynamic combined compressibility modulus for carbonates; total porosity for sandstones), and three physically motivated energy parameters: rotational mechanical power per unit bit area (Prot), axial crushing energy (AE/AEs), and hydraulic cleaning efficiency (Hce). Bit wear is quantified through a modified Hareland and Hoberock wear function requiring no laboratory measurements. Parameter selection used combined Pearson and Spearman correlation analysis across 16 candidate variables from a raw dataset of 9375 depth readings for Well A and 4443 for Well B (at 0.25 m intervals). The model was developed using nonlinear least squares regression (Levenberg–Marquardt algorithm) in MATLAB. Validated on two vertical wells penetrating mixed carbonate and clastic sequences in a Middle Eastern offshore field and benchmarked against four classical formulations, the model achieves R2 = 0.6568–0.6766 across full heterogeneous sections, improving on the best benchmark by margins of 0.36–0.46. Under lithology-specific calibration, R2 advances to 0.8239–0.9139, with MAPE reducing to 5.71%. The model is limited to two vertical wells in a single field; further field validation is recommended before broader deployment.