Machine learning based shear strength prediction in reinforced concrete beams using Levy flight enhanced decision trees.

Reinforced concrete (RC) T-beams are widely used in structural systems due to their efficient geometry and load-carrying capacity. However, accurately predicting their shear strength remains a challenge, particularly under complex loading scenarios. Conventional empirical approaches often struggle to adequately represent the complex and nonlinear relationships among structural design variables. In this study, a novel machine learning approach, termed Levy-DT, is introduced to enhance the prediction accuracy of shear strength in RC T-beams. The proposed method combines the structure of Decision Tree algorithm with Levy Flight, a stochastic optimization technique, to improve global search capabilities and avoid local minima. A comprehensive dataset comprising 195 experimentally tested T-beams is used to train and evaluate six different regression models, including optimized Decision Tree, Random Forest, AdaBoost, K-Nearest Neighbors, Ridge Regression, and the proposed Levy-DT. Model performance is assessed using multiple metrics such as R², RMSE, and MAE, with cross-validation employed for robustness. Systematic hyperparameter optimization is implemented for the baseline Decision Tree to ensure fair comparison. The results show that Levy-DT outperforms all other models, achieving the highest prediction accuracy with strong generalization. To further understand the model's decision-making process, SHAP analysis is carried out, identifying axial force and reinforcement depth as key contributors to the shear strength estimation. This study highlights the potential of integrating optimization techniques with machine learning for reliable and interpretable structural predictions.