AFFORDLEARN: EXPLAINABLE STACKED ENSEMBLE REGRESSION FOR INTERNATIONAL EDUCATION COST PREDICTION

Abstract

This paper provides a comprehensive discussion of machine learning regression models in predicting the total cost of international education, based on a real-life dataset that includes country, program, level, tuition, accommodation, visa, insurance, and features of living expenses. A common set of preprocessing was used to achieve uniformity between the models and an extensive selection of algorithms was examined including Linear Regression, Ridge, Lasso, ElasticNet, Decision Tree, Random Forest, Extra Trees, Gradient Boosting, XGBoost, and Multi-Layer Perceptron. Based on these standards, we introduce a new stacked ensemble model, AffordLearn, that combines XGBoost and Gradient Boosting as base learners and a Random Forest meta-learner. The suggested architecture represents a suitable combination of the boosting and bagging approaches to enhance the generalization and strength. As shown in experimental results (R2 = 0.9723, RMSE = 1.97, MAE = 1.46), the predictive accuracy of AffordLearn was highest compared to all the individual and ensemble baselines. The results emphasize the usefulness of hybrid ensemble learning in improving the modeling of the non-linear and complex interaction between costs, thereby providing a consistent and understandable predictor of international education costs. The research is informative to the students, policymakers, and other institutions that may want to use data-driven instruments to plan education costs globally.