การพยากรณ์ความเสี่ยงของการพ้นสภาพจากการเป็นนักศึกษาโดยใช้อัลกอริทึมการเรียนรู้ของเครื่อง: กรณีศึกษามหาวิทยาลัยเทคโนโลยีราชมงคล ล้านนา

Abstract

This independent study finds a predictive model for predicting at-risk students who will be academic dismissal. The predictive model will be identified for helping advisors to pinpoint students at risk so that the advisors can work more efficiently by convincing students concentrate on studying. It uses a machine learning technique that generates a predictive model from the Decision tree, k nearest neighbor, Logistic regression, naive bayes, Artificial neural network, Random Forest, Support vector machine algorithms and measures the performance of the algorithms for future use. Moreover, the factors that resulted in students being considered academic dismissal will be found by using SelectKBest, Correlation, Feature important, Forward Selection, Backward Elimination techniques to reduce the redundancy of the factors used in the predictive modeling. The sample used for prediction are prepared by using under sampling and oversampling techniques to enhance the learning efficiency of the forecasting model. The most efficient algorithm is the Support vector machine, which generates a predictive model using oversampling techniques. The efficiency was 8 8 .4 8 % and the key factors were (1) Student loan status (2) Parent relationship (3) Parent income (4) Province code Student address (5) Area (6) Family status (7) Previous GPA (8) Education level (9) Father income level (10) Parent living status (11) Graduation credits. The most important factor is student loan status, as these students are more willing to study than the average student, as they are required to repay the full amount with interest regardless of whether students graduate or fail to graduate. They are more committed to their studies than their counterparts.