C4.5, K-Nearest Neighbor, Naïve Bayes, and Random Forest Algorithms Comparison to Predict Students’ On Time Graduation

Abstract

Study program performance can be seen from the achievement of accreditation status, where one of the assessment instruments related to the graduate profile is the length of study. Graduation on time is one indicator of student’s success in obtaining a bachelor's degree and is an important attribute, because by being able to predict the period of study, universities can minimize student graduation failures by making more intensive planning, study escort, and guidance. Data mining classification techniques can be used to predict students graduation on time. Many data mining classification algorithms can be used, so it is necessary to make comparisons to determine the level of accuracy of each algorithm. The algorithms that will be compared in this study are C4.5, K-Nearest Neighbor, Naive Bayes, and Random Forest. The data used were 2,022 graduates from Informatics Engineering and Information System Study Program of STMIK Mikroskil Medan from 2011 to 2014, in which the attributes used include gender, regional origin, time of study, grade of Entrance Screening Examination, and Grade Point Average (GPA). The results of the classification process are evaluated using cross validation and confusion matrix to determine the most accurate data mining classification algorithm for predicting student graduation on time, where the K-Nearest Neighbor and Random Forest algorithms have the highest accuracy of 72,651%, followed by the C4.5 algorithm 72,453%, and the Naive Bayes algorithm 71,860%.