Optimizing Machine Learning with SSA and PSO for Anchor Bolt-Grout Bond Strength Prediction

The bond strength (τ) of the interface between the anchor bolt and grouting body (or rebar-concrete) is a key indicator used to evaluate the bearing capacity of anchorage engineering. And when rebars are subject to corrosion, τ also serves as an important durability metric. However, traditional experimental measurement of τ is complex, time-consuming and labor-intensive. In this study, based on pullout test data from 429 rebar-concrete specimens, we develop a machine learning method to construct a prediction model with strong generalization ability. Fundamental features-including specimen geometry, dimensions, material strengths, and corrosion rate-are used as inputs. The Sparrow Search Algorithm (SSA) and Particle Swarm Optimization (PSO) are used to fine-tune the hyperparameters of three machine learning models which are Random Forest (RF), Least Squares Boosting (LSBoost), and Generalized Additive Model (GAM). We perform a comparative error analysis of each model and benchmark them against three empirical formulas for τ. The unoptimized models exhibit low predictive accuracy and clear overfitting. After optimization using SSA and PSO algorithms, the prediction accuracy and overfitting issues are significantly improved, with the PSO-LSBoost model achieving the best performance (R² = 0.93). The PSO-LSBoost model's prediction accuracy for τ far exceeds that of the three empirical formulas. SHAP analysis reveals that the corrosion rate (C_w) contributes most to τ, while the rebar type (ST) contributes least. This work introduces a novel, efficient approach for predicting anchorage bond strength and assessing bolt durability, thereby enhancing the reliability of anchorage structures.