A Stacking-Based Meta-Ensemble Hybrid Artificial Intelligence Framework for Soil Shear Strength Parameters Prediction

Accurate estimation of soil shear strength parameters is essential for the safe and economical design of geotechnical structures; however, conventional laboratory testing is time-consuming, and individual machine learning models often exhibit limited generalization across heterogeneous soil conditions. This study proposes a stacking-based Meta-Ensemble Hybrid Artificial Intelligence framework for predicting cohesion (C) and angle of internal friction (ϕ) from routinely measured geotechnical properties. The proposed framework integrates ten complementary machine learning models through a meta-learning strategy to enhance prediction accuracy and robustness. A dataset comprising 854 laboratory-tested soil samples collected from six districts of Tamil Nadu, India, representing diverse soil classifications, was used for model development, while 20 independent samples were employed for external validation. Standard penetration test (SPT) N-value, bulk density, plasticity index, sand content, and fines content were selected as model inputs. The proposed framework achieved coefficients of determination (R2) of 0.9883 for cohesion and 0.9909 for the angle of internal friction for the testing set, better than conventional regression and individual machine learning models. The corresponding MAE and RMSE values were 4.00 and 9.5030 for cohesion, and 0.6595 and 1.5203 for the angle of internal friction, respectively. The trained model was further deployed as an intelligent decision-support application, MetEnSS2025, enabling rapid prediction of soil shear strength parameters for engineering practice. This study advances geotechnical engineering applications by employing artificial intelligence to predict soil shear strength parameters, supporting foundation design, soil characterization, and geotechnical decision-making while reducing reliance on extensive laboratory testing.