<![CDATA[On-time graduation is a key performance indicator for higher education institutions, including Universitas Padjadjaran, yet on-time graduation rates vary substantially across faculties (25.99%–87.80%). This study aims to build a predictive model of student on-time graduation and translate its results into managerial strategy recommendations. Using the CRISP-DM framework, six classification algorithms (Decision Tree, Random Forest, Naive Bayes, Logistic Regression, Gradient Boosted Tree, and XGBoost) were compared on academic and socioeconomic data from 25,872 students. The default (non-tuned) XGBoost model achieved the best performance (Accuracy 75.14%; AUC 0.832; F1-Score 76.82%), outperforming Logistic Regression and Gradient Boosted Tree, which both recorded 73.33% Accuracy. Feature importance analysis identified Faculty (35.76%) and Cumulative GPA (29.34%) as the two most dominant factors, jointly accounting for 65.10% of the total feature importance score. These dominant factors were then integrated into a SWOT analysis, yielding an IFAS score of 3.10 and an EFAS score of 2.95, placing Universitas Padjadjaran in Quadrant I (Growth) of the Internal-External Matrix. The TOWS matrix recommends integrating the XGBoost model into the university's integrated academic information system (SIAT) as an early warning system, prioritized for the five faculties with the lowest on-time graduation rates. This study demonstrates that integrating predictive analytics with strategic management can generate more targeted, evidence-based academic policy recommendations.]]>
Copyrights © 2026
