Using Machine Learning on Sales and Educational Data to Predict Student Behavior Turning Points: Academic or Financial Related

In a competitive educational market, it is crucial to predict the behavior of customers based on all available data sources. These sources can include both pre-enrollment and post-enrollment data. Although most existing studies rely solely on educational data collected after course enrollment, this study introduces a novel perspective by integrating pre-enrollment sales and marketing data, including user engagement during the acquisition phase, into predictive models. As predictive models, we used Random Forest, Gradient Boosting Trees, Artificial Neural Networks, and Logistic Regression models, along with feature selection and data analysis techniques. The purpose of this study is to determine the role of each data source in predicting key student events: absences from lessons, refund requests, and final project completion. Our findings show that sales data contribute the most to the prediction of money-related events, such as refund requests. Education data contribute the most to the prediction of education-related events, such as absence and the final project. The most important result is that education data does not improve the prediction of refund requests. The prediction of refund can be made solely on sales data even before the start of education. These insights can help educational companies improve retention strategies, improve financial forecasting, and optimize student support systems.