Deep Learning Neural Network for Statistical Modeling of Compressive Strength in Seybaplaya Bank Rocks: A Multivariate Analysis Incorporating Water Content, Porosity, and Density Parameters

Predicting uniaxial compressive strength (UCS) in geologically complex regions often poses significant challenges due to heterogeneous rock properties and limited field data. This study develops a site-specific, Bayesian-regularized neural network model to estimate UCS from three key parameters—moisture content, interconnected porosity, and real density—and implements, for the first time, a dedicated graphical user interface (GUI) tailored to the Bayesian Regularization Backpropagation algorithm. By leveraging locally acquired data from Seybaplaya’s carbonate–clay formations, the model captures the inherent variability of these bank rocks more effectively than conventional regression techniques. The approach employs a multivariate feedforward architecture with three hidden layers (10–6–1 neurons), trained under Bayesian regularization in MATLAB’s (trainbr) environment. This regularization framework not only mitigates overfitting but also promotes a parsimonious representation by constraining excessively large weight magnitudes. Extensive experimentation on a dataset of 134 samples—partitioned into training, validation, and test subsets—reveals a substantial reduction in mean squared error (MSE) (from ~29,100 to 326) alongside a marked decrease in sum of squared parameters (from 197 to ~7.63). These results underscore the network’s capacity to capture nonlinear interactions among the input features while maintaining balanced generalization across unseen data. Despite not reaching the stringent MSE target (1×10⁻⁷), the final model demonstrates moderate-to-strong predictive performance and highlights the significance of site-specific data in refining UCS estimations. This pioneering GUI-driven methodology provides an adaptable template for geomechanical analyses in regions typified by lithological heterogeneity and underscores the broader value of machine learning–based approaches for rock mechanics.