Machine Learning-based Predictive Analytics of University Enrollment
Grade Level at Time of Presentation
Junior
Major
Computer Science
Institution
Eastern Kentucky University
KY House District #
81
KY Senate District #
34
Faculty Advisor/ Mentor
Dae Wook (Wooky) Kim, PhD
Department
Department of Computer Science
Abstract
Machine Learning plays an indispensable role in a number of free or commercial predictive analytics tools used for multiple purposes, including web analysis, recommendation systems, biomarker discovery, and cyber security. Specifically, building a predictive model is highly desired for the purpose of university admissions. However, these tools raise significant robust concerns since the tools are not easy to customize the specific domain data and the university enrollment data is not an exception either. To tackle this research problem for the university enrollment data, we proposed a novel predictive system that effectively analyzes the enrolled applicants’ behaviors by integrating a collection of their behavioral patterns together, predicting if a prospective applicant is likely to enroll in a university or not. The extensive evaluation based on enrollment data collected from a database from Fall 2013 to Fall 2017 of a university within the U.S. has demonstrated that these patterns can accomplish high prediction rates of approximately 90% and low false positive rates of 5% using Random Forest as a statistic classifier and are also not sensitive to the selection of machine learning algorithms. In addition, we explored the relative importance of the proposed patterns, which has achieved the best prediction accuracy. Despite its high prediction accuracy, more enrolled applicants’ patterns could be discovered and incorporated into our system. Nevertheless, our system demonstrates the lower bound of the effectiveness of using enrollment patterns to predict prospective student matriculation for University admissions.
Machine Learning-based Predictive Analytics of University Enrollment
Machine Learning plays an indispensable role in a number of free or commercial predictive analytics tools used for multiple purposes, including web analysis, recommendation systems, biomarker discovery, and cyber security. Specifically, building a predictive model is highly desired for the purpose of university admissions. However, these tools raise significant robust concerns since the tools are not easy to customize the specific domain data and the university enrollment data is not an exception either. To tackle this research problem for the university enrollment data, we proposed a novel predictive system that effectively analyzes the enrolled applicants’ behaviors by integrating a collection of their behavioral patterns together, predicting if a prospective applicant is likely to enroll in a university or not. The extensive evaluation based on enrollment data collected from a database from Fall 2013 to Fall 2017 of a university within the U.S. has demonstrated that these patterns can accomplish high prediction rates of approximately 90% and low false positive rates of 5% using Random Forest as a statistic classifier and are also not sensitive to the selection of machine learning algorithms. In addition, we explored the relative importance of the proposed patterns, which has achieved the best prediction accuracy. Despite its high prediction accuracy, more enrolled applicants’ patterns could be discovered and incorporated into our system. Nevertheless, our system demonstrates the lower bound of the effectiveness of using enrollment patterns to predict prospective student matriculation for University admissions.