Comparing Data Mining Models in Academic Analytics

Introduction

High school graduates enroll in colleges to earn a college degree; however, some students do not graduate (Nara et al., 2005). An institution fails to retain its student if the student does not graduate from where they started. Seidman (2005) defines student retention as the “ability of a particular college or university to successfully graduate the students that initially enroll at that institution” (p.3). The U.S. Department of Education’s Center for Educational Statistics reported that only 50% of those who enroll in college earn a degree (Seidman, 2005). Noel and Levitz (2004) indicated that both private and public institutions have experienced escalating challenges associated with enrollment related issues in recent years. Student graduation is a very important display of academic performance and enrollment management to any university.

One of the concerns for a growing institution and its administration is the growth of the student population. Although university sets an aggressive goal for enrollment growth, there is still an underlying student graduation focus that the university has to keep in mind. That focus involves the ability of each student enrolled at the university to receive optimal educational opportunities and tools, leading to student graduation. An institution’s quality is assessed by its national ranking that consists of some factors like students with best grades, scholarships, students who do not leave and students who graduate.

The key to effectively understanding this complex balance between enrollment and graduation is in the application of statistical predictive models. Admissions personnel and management must be able to predict future criteria for a student who graduates or who does not graduate and be able to help students who will not graduate. Having such accurate predictions will greatly aid in the ability of the administration of a university to keep this positive balance between growth, quality, retention, and graduation. Predictive modeling for early identification of students at risk could be very beneficial in improving student graduation. Predictive models use data stored in institution databases that consist of student’s financial, demographical, and academic information. Predictive data mining therefore use large datasets to analyze student predictors of graduation. The predictive data mining decision planning is an innovative methodology that should be employed by universities. The heart of the data mining process involves building different predictive models and comparing to find the best model.

The purpose of this research study is to compare different data mining techniques as predictive models of student graduation. This study does not try to explore significant factors that contribute to student graduation rather compares the statistical predictive data mining models like logistic regression, decision tree, random forests and neural networks. The paper demonstrates all the cutting edge techniques in sampling, imputing, predictive models, and model comparison. Finally, this study will contribute to the meager research in effectiveness of data mining techniques applied in higher education and also help educational institutions better use data mining techniques to inform student graduation strategies. This study also used an ensemble classifier data mining technique called random forests that consists of many decision trees. Random forests have a very high accuracy in large datasets (Breiman, 2001), which has hardly been used in higher education data mining research. The significance of this study is in the discussion and comparison of several data mining techniques and their classification accuracy using important variables of student graduation.