Loading...
Utilization of AI to Predict Shear Strength Parameters of Soil Based on Their Physical Properties
AbdulGhafour, Husaen
AbdulGhafour, Husaen
Files
35.232-2025.46a Husaen AbdulGhafour.pdf
Adobe PDF, 3.56 MB
- Embargoed until 2027-01-15
Description
A Master of Science thesis in Civil Engineering by Husaen AbdulGhafour entitled, “Utilization of AI to Predict Shear Strength Parameters of Soil Based on Their Physical Properties”, submitted in November 2025. Thesis advisor is Dr. Mousa Attom. Soft copy is available (Thesis, Completion Certificate, Approval Signatures, and AUS Archives Consent Form).
Abstract
Shear strength is a fundamental geotechnical property that governs soil behavior under loading. Laboratory tests such as the direct shear test and the unconfined compression test are used to evaluate the shear strength parameters, namely cohesion, the angle of internal friction, and the Unconfined Compressive Strength (UCS) of soil before the design and construction of geotechnical structures such as foundations, retaining walls, slopes, and embankments. Although laboratory procedures are well-established and provide reliable measurements, they can be costly, time-consuming, and may not always be practical during early stages of geotechnical investigation. To address this issue, the main objective of this study is to develop machine learning–based predictive models to estimate cohesion, internal friction angle, and UCS based on simple index properties such as Atterberg limits, water content, dry density and grain size distribution to complement laboratory testing and assist engineers in early-stage geotechnical site investigations. Datasets were compiled from published studies and used to train eight machine learning algorithms to predict the target outputs, with hyperparameters tuned using either Optuna’s Tree-Structured Parzen Estimator (TPE) or RandomizedSearchCV with 10-fold cross-validation depending on each model. The results showed that the best performance was achieved by XGBoost for cohesion (R² = 0.738 and NRMSE = 11.02%) and UCS (R² = 0.931 and NRMSE = 6.27%), while the internal friction angle was most accurately predicted by the Multilayer Perceptron (MLP) neural network (R² = 0.884 and NRMSE = 5.66%). SHAP analysis and parametric evaluations conducted on the best performing models showed that cohesion is strongly influenced by Atterberg limits (especially the plastic limit), the internal friction angle depended primarily on clay and sand fractions, while water content was the most influential parameter for UCS prediction.