การคาดการณ์การออกกลางคันของนักศึกษามหาวิทยาลัยสงขลานครินทร์ด้วยเทคนิคการเรียนรู้ของเครื่อง

Abstract

Student retention rate plays a critical role and serves as an essential indicator of a tertiary institution’s success. However, not all first-time students complete their program at the same institution within a specified period of time: some students drop out of the program. Prince of Songkla University Hatyai Campus is no exception. From Academic Years 2013-2017, the student dropout rates rose by 19.18%. This research study adopted data mining and machine learning techniques to explore factors that predict the likelihood of a student dropping out, and to create a learning model of five-decision trees, which will be used for the prediction of the student dropouts. Data were collected from 33,930 students of Prince of Songkla University Hatyai Campus, from 6 intakes ranging from Academic Years 2015-2020, and with 39 variables. Collected data cover students’ learning achievements, students’ basic information, and students’ family background. Data were classified into two categories: Undergraduate and Postgraduate. As for undergraduate category, the study found that Light Gradient Boosting Machine is the most appropriate methodology, as it yielded the highest value of the area under the curve of 93.03%, and the accuracy value of 89.99%. The top factors that predict the likelihood for student dropouts include the accumulated (overall) grade point average (GPAX); academic year; Grade Point Average (GPA); semester; pre-university GPAX; and pre-university English scores, respectively. As for postgraduate category, the study found that the Random Forest is the most appropriate methodology, as it yielded the highest value of the area under the curve of 78.86%, and the accuracy value of 85.28%. The top factors that predict the likelihood for student dropouts include GPAX; academic year; semester; social and humanity science; GPA; supplementary class; and Plan A, A2-Type, respectively. In the final procedure, the researcher implemented the obtained models for making a prediction with the actual data, and visually presented the results of the analysis in the dashboard report, which can be used for monitoring possible risks. This will enable respective staff to give immediate assistance to the students who are in needs or show the likelihood to drop out, and help the management board in making decisions and devising management plans to minimize the dropout rate in their institution.