Exploring the impact resistance performance of RC beams based on an enhanced interpretable automated machine learning approach

The analysis and improvement of the maximum displacement of reinforced concrete (RC) beams under impact loads are pivotal in augmenting their impact resistance. The present study introduces an enhanced interpretable automated machine learning (EI-AutoML) approach for accurately predicting the mid-span deflection of RC simply supported beams subjected to impact loads. The method aims to mitigate the substantial time and technical requirements linked to conventional Machine Learning (ML) model development. Enhancements in model execution efficiency were achieved by fine-tuning parameters and evaluation metrics using the Tree-based Pipeline Optimization Tool (TPOT), along with the simultaneous execution of multiple TPOT instances. The enhanced TPOT algorithm generated the optimal stacking model in the experimental group, while several commonly employed traditional machine learning models were used as the control group. The results show that the enhanced TPOT's stacking model demonstrated outstanding robustness and generalization capabilities, with evaluation metrics significantly outperforming traditional ML models. The optimal model achieved an R2 of 0.975 and a Root Mean Square Error (RMSE) of just 3.688 mm. Moreover, three interpretative analysis methods, such as SHapley Additive exPlanation (SHAP) with five-fold cross-validation, were used for feature importance and parameter analysis. It also examines the interactive effects of various features on deflection, providing a theoretical foundation for optimization in engineering practice. To apply it in practical engineering scenarios, a symbolic regression analysis was performed to derive empirical formulas based on this model. Additionally, a Graphical User Interface (GUI) was created to enhance its usability in engineering applications.