머신러닝을 활용한 한계대학 예측 요인 탐색 연구

Abstract

This study seeks to find the criteria for preparing government and social countermeasures by using machine learning to explore predictive factors for universities in business crisis. Between 2012 and 2020, universities with low evaluation results in the government's university restructuring evaluation are defined as universities in business crisis. The goal was to discover factors that could predict a university in business crisis by using data such as students, faculty, and financial data, which are university components. As a result of the analysis, when using the Random Forest, it was possible to accurately identify universities in business crisis compared to the Logistic Regression model. The results of predicting universities in business crisis through shap analysis and statistical (t-test) are as follows. First of all, the ROC-AUC value that can evaluate the predictive power was checked by applying the logistic regression model and random forest to the same data for the prediction of universities in business crisis, and the logistic regression model was evaluated as 0.819 for universities and 0.854 for colleges. , Random Forest was evaluated as 0.927for universities and 0.943 for colleges. In the case of universities, the factors that have a positive(+) influence on the prediction of universities in business crisis are the dropout rate, region_Gyeongnam. On the other hand, the factors that have a negative(-) effect on the prediction of universities in business crisis are education expenses per student, education expense reimbursement rate, enrollment rate, maintenance recruitment rate, new student recruitment rate, employment rate, operating expense burden rate, and research expenses per full-time faculty member. In the case of colleges, the factors that have a positive (+) influence on the prediction of universities in business crisis are the carryover rate, teacher retention rate, region_Gangwon, dropout rate and debt ratio. On the other hand, the factors that have a negative(-) effect on the prediction of universities in business crisis are education expenses per student, student recruitment rate, university size, education expense reimbursement rate, maintenance recruitment rate, operating expense burden rate, new student recruitment rate, scholarship payment rate, and research expenses per full-time faculty member. The contribution of this study is to propose a new methodology to predict universities in business crisis using machine learning techniques, which are widely used recently.