<p>This study presents a data-driven machine learning (ML) based method to assess the seismic vulnerability of reinforced concrete (RC) educational buildings. The appropriate features are key to developing an optimal model in predicting target variables, Storey Shear Ratio (SSR), a key risk indicator, and analytical response in seismic vulnerability assessment. The feature selection methods and correlation matrix can be used to identify the most influential features. This study implemented four feature selection. Filter Methods, Wrapper Methods, Embedded Methods, and Bayesian Methods, to identify the most influential parameters in predicting the SSR. The analyses are conducted with two sets of parameters, 19 parameters and 8 parameters, comprising 268 RC educational building datasets. Pearson’s correlation analyses are also conducted to avoid multi-collinearity on both datasets to assess relationships between features and SSR. Feature importance scores were assessed and compared among methods, and correlation analysis was performed to identify multi-collinearity features and validate their relevance to SSR. The Bayesian Method showed limitations in showing feature importance due to its probabilistic nature, whereas other methods yielded consistent feature ranking. The results highlighted that construction year, building condition, number of stories, typical floor area, redundancy, pounding, additions, and structural irregularity were consistently ranked and strongly correlated as highly influential features. The reduced 8 parameters models retained the most influential features, implying the potential for comparable performance to the complete set of 19-parameter models, supporting its capability in ML-based rapid seismic vulnerability assessment frameworks. This study could enrich the interpretability and efficiency, reduce overfitting of ML-based seismic vulnerability assessment, and provide a quick methodology for risk evaluation in educational RC buildings.</p>