Abstract

Student attrition within schooling systems represents a persistent obstacle to both individual progress and broader societal improvement. This research presents a predictive model set to notify teachers of students most likely to drop out of school early, leveraging ensemble learning methods on a dense, multi-dimensional data set. The data portal consists of socio-demographic and educational aspects: residential area, first language, home occupation and level of education, number of family members, school distance, age, gender, level of education of mother, grade level, mode of transport to school, and number of siblings. Together, these measures chart the path to the dropout outcome. An ensemble algorithm suite of Random Forest, XGBoost, and Stacking Classifier leverages their capacity to capture complex, non-linear relationships, thus raising predictive accuracy. Model performance is evaluated by Accuracy, Recall, F1-Score, and ROC-AUC. Results indicate a consistent superiority of the ensemble techniques over standard algorithms producing actionable intelligence for teachers, school administrators, and policymakers. This question informs the build-out of future-oriented, evidence-based warning systems designed to reduce dropout rates and improve overall school performance.

Authors

Rupali Ambalal Jadhav, Rupal Parekh
Atmiya University, India

Keywords

Educational data mining, Dropout, Ensemble learning, Early Warning System, School Education