A multifaceted data mining approach to understanding what factors lead college students to persist and graduate

Universities in the United States are facing the serious issue of high dropout rate and low graduation rate of four-year college students. This paper describes a host of data mining approaches to help tackle this issue. Specifically, we utilize the following approaches to identify factors that contribute to student persistence and graduation: (1) a visual analysis to identify bivariate relationships and to understand the flow of students in an educational institute; (2) an ensemble feature selection method to recognize factors that have a significant impact on a student's persistence and graduation; (3) classification and prediction algorithms to predict whether a student will persist in a given semester and ultimately graduate; and (4) a variety of association patterns to help education practitioners gain further insights into factors that affect persistence and graduation. To evaluate the above approaches, we use data originated from a local academic program. Our analyses have resulted in both interpretable and actionable outcomes. For example, the ELM (Entry Level Mathematics) score was identified as one of the most influential factors in predicting a student's third-term persistence, and furthermore graduation. This insight has in turn motivated the above program to enroll their students with low ELM scores in a remedial math course before they start their freshmen year. Among the classification algorithms under consideration in this study, we have demonstrated that Naïve Bayesian is more suitable for predicting graduation, whereas AdaBoost and SVM are better at predicting persistence.