PREDICTIVE ANALYTICS FOR TEACHER ORGANIZATIONAL CITIZENSHIP BEHAVIOR: EVALUATING MAXIMUM-MARGIN AND ENSEMBLE TREE ARCHITECTURES IN EDUCATIONAL MANAGEMENT

Abstrak

In this paper, we propose a pure data-driven predictive analytics framework to classify organizational citizenship behavior (OCB) of teachers in 226 teachers, which addresses the critical lack of complex modeling in traditional school management research. It is important to study these dynamics because private vocational schools often suffer from severe resource constraints and fluctuating institutional support, and the survival of the institution depends heavily on teachers’ voluntary willingness to contribute beyond formal job boundaries. The findings of this study indicate that the Support Vector Machine (SVM) model with an RBF kernel provides the best predictive performance. We tested three supervised machine learning algorithms with kuesioner data on an 80:20 split with an optimal Accuracy of 73.91% and an excellent classification ROC AUC Score of 80.91%. This beats the Random Forest Classifier (69.57% Accuracy, 78.64% ROC AUC) and the Gradient Boosting Classifier (69.57% Accuracy, 76.75% ROC AUC). Furthermore, our feature importance evaluations consistently show that School Climate (Z1) is the most important anchor and sensitive predictor of teacher voluntarism, followed by Principals’ Emotional Support (X1) and Recognition and Appreciation (X2). The values of local cultural wisdom such as Gotong Royong (X3) and Honesty Values (X4) are the highly stable baseline structural predictors within the non-linear decision boundaries. Future work should extend this predictive framework by incorporating advanced deep learning models and longitudinal behavior tracking across a diverse array of public and private school networks to further refine data-driven educational policymaking