A declining school-age population and rising university dropout rates threaten the stability and sustainability of higher education. This study aims to empirically classify dropout risk groups, describe their characteristics, and examine associated predictors. Using data from 2,174 students identified as at risk by University A’s machine-learning based dropout prediction model, K-means clustering identified three distinct dropout risk groups: early involuntary dropout risk, mid-term dropout risk, and early latent dropout risk. Random Forest analysis determined key predictors distinguishing the clusters, and multinomial logistic regression assessed the relative influence of these variables. Course performance, attendance, counseling patterns, number of scholarship awards, academic changes, and commuting distance emerged as primary discriminators among risk groups. Findings indicate that dropout risk extends beyond academic underperformance and reflects the combined effects of academic, behavioral, economic, and institutional factors. By integrating cluster analysis with machine learning, this study offers an evidence-based framework for early risk detection and informs targeted student support strategies in higher education.