Comparative Evaluation of Feed-Forward Neural Networks for Predicting Uniaxial Compressive Strength of Seybaplaya Carbonate Rock Cores

Accurate estimation of uniaxial compressive strength (UCS) of carbonate rocks is critical for the design and stability assessment of geotechnical structures in karst‐influenced environments. This study evaluates four feed-forward artificial neural network (ANN) architectures as Radial Basis Function (RBF), Bayesian Regularized (BR), Scaled Conjugate Gradient (SCG), and Levenberg–Marquardt (LM) to predict UCS from three readily measured inputs: water content, interconnected porosity, and real density. A dataset of 50 core specimens from the Seybaplaya quarry in Campeche, Mexico, was partitioned into training and testing subsets under consistent preprocessing. Model performance was assessed via mean absolute error (MAE), root mean squared error (RMSE), mean absolute percentage error (MAPE), and coefficient of determination (R²), averaged over 30 independent runs. Statistical significance of differences in test-set mean squared error (MSE) was examined using the Friedman test with Benjamini–Hochberg correction. Sensitivity analysis based on partial derivatives quantified the relative influence of each input feature.The RBF network achieved the highest predictive accuracy (median R² = 0.975; RMSE = 1.313 MPa) and significantly outperformed BR and SCG models (adjusted p < 0.01). Bayesian regularization offered robust generalization (MAE = 1.164 MPa; R² = 0.967), while SCG and LM converged faster but with slightly lower accuracy. Sensitivity scores indicated interconnected porosity (54.4 %) as the dominant driver of UCS predictions, followed by water content (30.9 %) and real density (14.7 %). These findings demonstrate that RBF-based ANNs, combined with appropriate regularization and sensitivity assessment, provide a reliable framework for UCS prediction in heterogeneous carbonate formations.