Machine-Learning Models to Correlate Design and CPT Parameters

Many geotechnical models have been developed to predict soil behavior, but they often struggle to create accurate links between soil design parameters and geotechnical investigations. This leads to low prediction accuracy, which causes technical issues, higher construction costs, and project delays. In cohesive soils, the undrained shear strength (Su) is a key factor affecting several geotechnical applications. This study uses machine learning (ML) models to predict Su based on cone penetration test (CPT) data. Two ML models, Bayesian Deep Neural Networks (BDNN) and Support Vector Machine Regression (SVMR), were used with nine input variables for point-estimated predictions. Their performance was assessed using multiple metrics, including mean absolute error (MAE), mean squared error (MSE), R-squared score, variance accounted for (VAF), and the a۲۰-index. The SVMR model showed strong reliability, effectively capturing soil variability with an R² of ۰.۸۹۸ for the training dataset and ۰.۸۵۰ for the testing dataset. These results highlight the potential of ML methods to improve geotechnical predictions while keeping costs down.

کلیدواژه ها:

Machine Learning ، Cone Penetration Test ، BDNN ، SVMR