Optimized CatBoost-Based Soft-Computing Models for Prediction of the Ultimate Bearing Capacity of T-Shaped Footings Subjected to Eccentric Load

This study introduces a novel approach employing four innovative hybrid CatBoost-based soft-computing models to accurately predict the ultimate bearing capacity of T-shaped strip footings on cohesive-frictional soils. This research aims to enhance the design's safety and economy by defining critical parameters such as the footing width (B), embedment length (D), load eccentricity (e), soil unit weight (γ), and shear strength parameters (cohesion: c, and internal friction angle: \(\phi\) $\varphi$ ). While prior studies have explored various methods for predicting bearing capacity, this research leverages advanced hybrid algorithms to improve accuracy and reliability. Data collection and processing involved deriving plastic solutions for the bearing capacity factor of eccentrically loaded footings through upper and lower bound finite element limit analysis (FELA) under 2D plane strain conditions using OptumG2 software. The bearing capacity factor (N) was determined by considering the following dimensionless parameters: the eccentric length and width ratio (e/B), embedment length and width ratio (D/B), soil strength factor (γB/c), and internal friction angle ( \(\phi\) $\varphi$ ). The key findings indicate that all four hybrid models produced highly accurate predictive models. Among these models, the FPA-CatBoost model demonstrated exceptional accuracy without overfitting. The study also proposed FELA results as charts illustrating the bearing capacity factor and failure mechanisms of T-shaped strip footings in cohesive-frictional soils and included a comparative analysis with findings from previous literature. These results underscore the potential of hybrid CatBoost models in providing precise predictions, ultimately leading to safer and more economical foundation designs.