Predictive modeling of shear strength in fly ash-stabilized clayey soils using artificial neural networks and support vector regression

This study explores the prediction of shear strength in fly ash-stabilized clayey soil using Artificial Neural Network (ANN) and Support Vector Regression (SVR). Clayey soils, characterized by low shear strength and high plasticity, present significant challenges in construction, necessitating effective stabilization methods. Fly ash, a byproduct of coal combustion, provides a sustainable alternative due to its pozzolanic properties. The research integrates ANN and SVR to model complex relationships between soil properties (grain size distribution, plasticity index, liquid limit, plastic limit, moisture content), fly ash content, and curing periods. Laboratory experiments and triaxial shear tests generated the dataset for training and testing the models. The ANN model achieved a training R² of 0.93 and a Mean Squared Error (MSE) of 0.00, while the testing R² was 0.69 with an MSE of 0.01. In contrast, the SVR model outperformed ANN with a training R² of 0.95 and MSE of 0.01, and a testing R² of 0.83 and MSE of 0.00. Sensitivity analysis identified key factors influencing shear strength predictions, with SVR demonstrating superior generalization capabilities. The study concludes that SVR is a more reliable tool for predicting shear strength in stabilized soils, contributing to sustainable construction practices.