Predictive Analytic for Student Dropout in Undergraduate using Data Mining Technique

Abstract

The purposes of this research were 1) to analyze the factors that involved with the dropout of undergraduate students 2) to propose a model for predicting the dropout of undergraduate synthesize 3) to compare the performance of 3 different classification techniques, including Decision Tree, K-Nearest Neighbors, and Naive algorithms. The data was collected from the undergraduate student’s registration database of Ubon Ratchathani Rajabhat University during the academic years from 2015 to 2017. The dataset has 11 attributes and 13,729 records. The data were analyzed using the Information theory selection method. The results showed that 1) there are 8 factors that influencing student’s dropout 2) Those factors were used to build models with the different techniques, Moreover, the cross-validation with 10 folds method was used to evaluate the best prediction accuracy of each technique. 3) the result suggested that the Naive Bayes model has the best performance among all techniques. It has the average accuracy of 93.58 %, which are higher than Decision tree and K-Nearest Neighbors which have the average accuracy of 93.52 % and 87.95 %, accordingly. The findings also indicated that students’ decision to dropout was significantly influenced by the student loan, major of study, grade point average, and the occupation of their parents.