Robust Tree-Based Machine Learning Algorithms for Predicting Drag Anchor Performance

Drag embedment anchors (DEAs) are anchors that generate holding capacity by embedding into the seabed under tension. DEAs are widely used in offshore mooring. This study presents a data-driven framework using decision tree regression (DTR) and random forest regression (RFR) to predict DEA holding capacity and anchoring efficiency in clay seabed. The DTR and RFR models were validated using experimental measurements. By leveraging geotechnical and geometric parameters such as undrained shear strength, fluke–shank angle, and anchor weight, the models captured nonlinear relationships and provided robust predictions. Multiple analyses were conducted to determine the most influential parameters and the optimal machine learning (ML) models. The sensitivity results showed that anchor weight and fluke length had the greatest impact on holding capacity, whereas soil unit weight and the bearing capacity factor were the dominant variables affecting efficiency predictions. RFR models outperform DTR in accuracy and generalization, with RFR 1 achieving the highest predictive reliability. These findings advance geotechnical engineering practices, enabling optimized DEA designs for offshore energy infrastructure.