Data-Driven Preliminary Seismic Evaluation of School Gymnasiums: An Integrated Framework of Artificial Intelligence and Multivariate Statistical Analysis

School gymnasiums in Taiwan often serve as emergency shelters, demanding vital seismic evaluation. Conventional collapse ground acceleration (Ac) assessment via finite element software requires exhaustive computational time and detailed reinforcement data unavailable during early-stage planning. To bridge this operational gap, this study develops a data-driven framework for preliminary seismic evaluation. This framework offers structural designers or architects a rapid planning tool for prospective facilities and enables regional screening to identify vulnerable existing gymnasiums, complementing high-fidelity modeling. An evolutionary support vector machine inference system (ESIS) is proposed to efficiently predict Ac utilizing basic geometric and layout parameters. Concurrently, gene expression programming derives explicit equations to enhance engineering interpretability, while multivariate statistical analyses, including principal component analysis and cluster analysis, together with correlation analysis, are incorporated to investigate seismic characteristics statistically and identify dominant seismic factors. Based on an empirical inventory of 479 standalone gymnasiums, results reveal that dominant seismic factors vary distinctly among structural clusters. Compared with several baseline AI models, the ESIS achieves superior generalization performance, yielding RMSE values of 0.067–0.086 and R2 values of 0.794–0.918 on testing datasets. Ultimately, this integrated framework provides an efficient and highly accurate decision-support tool for macro-scale screening and preliminary seismic performance assessment.