Explicit classification for seismic failure modes of corroded reinforced concrete columns using class‐weight‐optimized logistic regression

Abstract Accurately classifying seismic failure modes of corroded reinforced concrete (RC) columns is critical for safety assessment but is hindered by severe class imbalance and limited interpretability of existing models. This study proposes an interpretable classification framework based on class‐weight‐optimized logistic regression (CWO‐LR). A cost‐sensitive learning strategy is introduced by optimizing class‐specific weights through Bayesian optimization with cross‐validation. The optimized model is incorporated into a two‐stage hierarchical scheme to first distinguish ductile and brittle failures, and then identify brittle modes. The method is validated using 289 experimental tests of corroded RC columns and compared with traditional logistic regression (LR) and explicit classification models. Results show that CWO‐LR provides explicit classification formulas while effectively mitigating class imbalance. Compared with traditional LR, it improves overall accuracy by about 12% and significantly enhances the F1‐scores of minority failure modes. The model also demonstrates stable performance under varying imbalance ratios, indicating good generalization. This study offers a reliable and practical tool for seismic performance evaluation of deteriorated RC columns.