摘要：Students’ withdrawal problem is one of the main concentration of enrollment management at educational institutions as it negatively affects their performance and reputation. This paper discusses two types of students’ withdrawals which includes long-term dropout and the short-term dropout and considers this problem as a multi-class classification problem rather than a binary classification problem. We first introduce a novel (RG*) method to generate ruleset using multiple rules learning classifiers including Decision Trees and Rule induction methods to improve the accuracy and interpretability of the classification. Then we propose a predictive framework based on the RG* to predict at-risk students and to address students’ data problems such as imbalanced and high-dimensionality. Two groups of criteria are used to evaluate the proposed framework including: model performance and interpretability. The results revealed the possibility of a tradeoff between the performance and interpretability of the classification outputs through exploiting the ability of the multiple classifiers. In addition, the proposed framework shows a significant improvement in predicting both dropout and stopout students' compared with using individual classifiers.
