<![CDATA[Student attrition remains a persistent problem in Indonesian higher education, but institutional decision-support tools rarely combine time-to-event modeling with longitudinal financial data. Prior dropout studies rely on single-point-in-time machine-learning classifiers such as Random Forest, XGBoost, and CatBoost, which ignore attrition timing and treat financial status as static, while existing multilevel survival studies capture program-level heterogeneity without individual financial trajectories. This study proposes the Student Financial Trajectory Index (SFTI), a semester-level composite of payment ratio, arrears ratio, arrears trend, and cumulative late-payment fines, and integrates it into a multilevel discrete-time survival analysis. Administrative panel data from 2,528 students in the 2021/2022 through 2025/2026 cohorts, across seven study programs at a private Indonesian university, were restructured into a person-period dataset. Dropout was defined both administratively and operationally (two consecutive inactive semesters), with the study program as a random-intercept cluster. Preliminary analysis shows that students in the higher-risk SFTI group have a dropout hazard of 0.51% per semester at risk, compared with 0.08% in the lower-risk group, suggesting that financial trajectory carries an early-warning signal beyond academic performance. Formal multilevel modeling shows academic predictors (GPA and cumulative leave) as the dominant predictors of dropout hazard: the SFTI coefficient in the combined model was positive but not statistically significant (? = 0.10, SE = 0.41, p = 0.80). Model B achieved a marginal QIC improvement over the academic-only model (95.246 vs 95.255) and identical held-out performance (AUC = 0.960; Brier = 0.002). The final model produces a semester-by-semester early-warning risk score to support institutional decision-making.]]>
