Predicting the Timeliness of Student Graduation Using Decision Tree C4.5 Algorithm in Universitas Advent Indonesia

Abstract

The purpose of this study is to predict whether a student will graduate on time or not, using the data collected from the database of Academic Administarion Bureau of Universitas Advent Indonesia (UNAI) for the academic period of 2009–2013. There are 9 (nine) attributes that were used to predict the timeliness factors, and the method used for predicition is the C4.5 decision tree. This study also used SMOTE (Synthetic Minority Oversampling Technique) in the WEKA application to balanced data in minor class. The result shows that the attribute GPA, was the highest root that gave the highest influence. Another attributes that also has influence in predicting the student s graduation timelines are: repeating courses, study leave, and gender, while the attribute religion, is the attributes that are less influential. The accuracy on cross validation 10 Folds with SMOTE and without SMOTE the results shows that: with SMOTE 83,055% and without SMOTE 82,644%. Split Test 70:30 with SMOTE and without SMOTE, the results are 82,026% and 84,015%. The use of SMOTE increases the value of precision and recall. The value of precision for not ontime student, with SMOTE 82.5% and without SMOTE 76.4%, the value of recall with SMOTE 80.6% and without SMOTE 61.9%.