A Statistical Modelling for Dropout Patterns of College Students : Focusing on H University

This study was conducted to analyze dropout patterns of H University students and construct a prediction model for churning of students. Keeping a high enrollment rate of students in a local university is very important for its sustainability and survival, which forced us to develop a good explainable predictive model for dropout probabilities of each student. We applied a mixed effect logistic regression model in order to predict a dropout probability based on 11 explanatory variables such as admission type of processes, first semester grades, total admission scores, category of high schools, and departments of application, etc. The constructed model tells us that department of application is a significant variable as a random effect on which fixed effects for total admission scores and first semester grades should be interpreted differently. The main finding is that in each department the proposed model was able to discriminate a high-probabilistic dropout student although they have showed a good performance in total admission scores and first semester grades. It is highly expected that our model can be effectively applied to manage the dropout rates and enhance the enrollment rates of the students.