An Advanced Approach for indirect estimation of Shear Strength Parameters from Easily Obtainable Soil Properties Across Diverse Soil Types Using Machine Learning Models

Proper estimation of the soil shear strength parameters is a prerequisite to characterize subsurface behavior when modeling the Earth system and working in geotechnical engineering fields. This work implies a machine learning-based method for predicting soil shear strength parameters, cohesion, and internal friction angle, based on only fundamental properties of soil composition (wight percentage of clay, silt, sand, and gravel). The thrust is to design and validate a simultaneous artificial neural network model that can estimate both parameters at the same time and thus respect their actual correlation. 95 soil samples of different sorts tested with direct shearing served in training and validation of the models. Simultaneous ANN model performance was compared systematically with two other approaches: (1) using two separate ANN models for the two parameters, and (2) using an MLR model improved by PCA to counter multicollinearity issues. The models were evaluated employing some statistical indicators: R², RMSE, MAE, CV-RMSE, and a20-index. The simultaneous ANN model performed far better than the other two methods with R² values of 0.95 and 0.90 for cohesion and internal friction angle, respectively. This outperformance in terms of accuracy and efficiency substantiates the advantage offered by joint modeling of properties that vary together. This signifies the promise offered by data-driven modeling for subsurface parameter estimation that is reliable, thus creating a cost-effective and scalable tool catering to early-stage geotechnical modeling and Earth system simulations.