Predicting the Risk of Course Change in Online Education Using the Causal Shapley Method

The proliferation of online education has generated new demands for improving the quality of education through an understanding of student behavior patterns. This study introduces an approach to elucidate the complex relationship between student behavior and educational outcomes, specifically focusing on predicting course changes in online schools. To enhance the interpretability of the predictive model, we introduced the Causal Shapley value, which incorporates causal inference, in comparison to traditional SHAP values, to assess the impact of features. This approach confirmed that attendance rate and the number of students per campus are significant features for predicting course changes. The Causal Shapley value provided new insights, especially highlighting the number of students per campus as a causal factor in dropout predictions, offering a finer perspective on the impact of features than traditional SHAP values. This methodology goes beyond mere correlations, allowing for a deeper understanding of how features influence prediction outcomes, thereby providing vital information for educational institutions to design intervention strategies tailored to the individual needs of students.