What question did this study set out to answer?

This research aims to enhance the prediction of undrained shear strength (USS) in cohesive soils using a hybrid machine-learning approach.

May 2, 2026SHILAP Revista de lepidopterología0 citationsOpen Access

Prediction of Undrained Shear Strength of Soil Using Hybrid Automated Radial Basis Function

MAMahsa Heydari Alikamar

Key Points

This research aims to enhance the prediction of undrained shear strength (USS) in cohesive soils using a hybrid machine-learning approach.
Developed a hybrid model combining Radial Basis Function with Smell Agent Optimization, Motion-Encoded Particle Swarm Optimization, and Seagull Optimization Algorithm.
Trained models using 70% of available dataset and tested on an independent 30% subset to assess generalization performance.
Evaluated model accuracy with statistical indicators like R² and CV_RMSE.
The hybrid RBF model achieved a maximum R² value of 0.993 during testing.
CV_RMSE of the RBF model was recorded at 0.053, indicating high accuracy.
Demonstrated effectiveness of meta-heuristic techniques in enhancing predictive capabilities of RBF models.

Abstract

Accurate estimation of undrained shear strength (USS) is essential in geotechnical engineering applications, including foundation design, bearing capacity evaluation, and slope stability analysis, as it directly governs the short-term mechanical behavior of cohesive soils under undrained conditions. Conventional laboratory and in-situ tests, such as the standard dilatometer test (DMT) and cone penetration test (CPT), are commonly used for this purpose but are often associated with high cost, time consumption, and uncertainty due to testing assumptions, soil disturbance, equipment limitations, and variable field conditions. These limitations restrict their applicability, especially in large-scale or complex projects, highlighting the need for reliable predictive alternatives. This study proposes a hybrid machine-learning framework for predicting USS using a Radial Basis Function (RBF) combined with Smell Agent Optimization (SAO), Motion-Encoded Particle Swarm Optimization (MPSO), and the Seagull Optimization Algorithm (SOA). The hybridization aims to enhance the learning capability of the RBF model by optimizing its internal parameters and improving convergence behavior. The developed models were trained using 70% of the available dataset and evaluated on an independent 30% testing subset to ensure robust generalization performance. Model accuracy and reliability were assessed using multiple statistical indicators, including the coefficient of determination (R²), coefficient of variation of root mean square error (CVRMSE), Theil’s inequality coefficient (TIC), weighted mean absolute percentage error (WMAPE), and Nash–Sutcliffe efficiency (NSE). The results indicate that the hybrid RBSO model achieved the highest prediction accuracy, with an R² value of 0. 993 and a CVRMSE of 0. 053 in the testing phase. These findings highlight the effectiveness of meta-heuristic optimization techniques in improving RBF-based models and demonstrate their potential for reliable USS prediction in practical geotechnical engineering applications.

AI에게 질문

Bookmark

View Full Paper