<![CDATA[ Higher education institutions are required to produce graduates who are competitive in the job market, yet tracer study data is often analyzed only descriptively, leaving deeper patterns unexplored. This study proposes an integrated approach combining Random Forest, SHAP (SHapley Additive exPlanations), and fairness evaluation to analyze alumni career outcomes. Data were sourced from the 2024 tracer study of Universitas Royal, comprising 466 alumni records with a binary target variable indicating employment status. Preprocessing included missing value handling, one-hot encoding, and stratified train-test splitting with an 80:20 ratio. Class imbalance was addressed using the class_weight='balanced' parameter. The Random Forest model achieved an accuracy of 95.7%, F1-score of 0.946, and AUC-ROC of 0.938, significantly outperforming the dummy baseline of 58.5%. Cross-validation across five folds yielded a mean accuracy of 96.4%, confirming model stability. SHAP analysis revealed that features related to educational level appropriateness and relevance of study field to current occupation were the most dominant predictors. Fairness evaluation per study program showed minimal performance disparity between Sistem Informasi (94.7%) and Sistem Komputer (100%), indicating no significant predictive bias. This framework demonstrates that explainable and fair machine learning can serve as a reliable foundation for evidence-based academic policy making]]>
Copyrights © 2026
