Interpretable multi-index seismic assessment framework for reinforced concrete columns using voting ensemble methodology

Abstract

Reinforced concrete (RC) columns constitute critical load-bearing components that determine structural safety under seismic actions. While conventional assessment methods relying on empirical formulas struggle to address complex multi-parameter interactions, this study proposes an interpretable Voting-based Optimization Tree Regression (VOTR) framework for multi-index seismic performance evaluation. A rigorously validated database of 145 experimental specimens was established, with data rationality confirmed through Pearson correlation and mutual information analysis. The VOTR model integrates Bayesian hyperparameter optimization with 10-fold cross-validation, demonstrating enhanced predictive accuracy over baseline machine learning methods in estimating ductility coefficients (Test R2 = 83.76 %) and yield displacements (Test R2 = 91.26 %). Through the analysis of data correlation and the significance of the SHAP model characteristics, it is found that the axial compression ratio, hoop ratio, and shear span ratio are the critical factors influencing ductility and yield displacement. The interpretability analysis bridges data-driven predictions with mechanical principles, revealing quantitative relationships between design parameters and seismic performance metrics. Comparative validations confirm the model's effectiveness in supporting performance-based seismic design optimization. This framework advances traditional prescriptive methods by providing a transparent, physics-informed assessment tool that balances computational accuracy with engineering interpretability for RC column safety evaluations.