Prediction of non-uniform shrinkage of steel-concrete composite slabs based on explainable ensemble machine learning model

Abstract

The non-uniform shrinkage (NUS) of steel-concrete composite slabs has been proved to significantly affect the long-term performance of composite slabs. But the reliable analysis method has not yet been proposed due to the multi-factor coupling mechanism of material composition and boundary conditions. A database containing 782 samples was established in this paper, and the index set containing 5 feature parameters was selected by analyzing the physical mechanism of NUS of steel-concrete composite slabs. The predictive capabilities of 8 machine learning (ML) models with different working mechanisms were compared, and the reliability of ML models was verified by Taylor and box scatter from the perspective of global and single sample, respectively. Furthermore, sensitivity analysis was conducted using the SHapley Additive exPlanations (SHAP) method to evaluate the impact of each parameter on the NUS of steel-concrete composite slabs. Finally, a user-friendly graphical user interface (GUI) was developed to guide engineering design. The limitations and potential improvements of the ML models established in this paper were discussed. The results indicate that the variance inflation factor (VIF) method and correlation analysis can effectively mitigate the influence of data multicollinearity and variable collinearity, and the pearson coefficient and VIF value of the input variables are less than 0.35 and 5. Gradient boosting decision tree (GBDT) has the best prediction performance, with R2 of 0.933 and 0.927 on the training set and testing set, respectively. Relative distance (x) and age (t) are the key parameters affecting the NUS of steel-concrete composite slabs. The procedure and GUI developed in this paper can effectively guide the intelligent prediction of the composite slab NUS.