Universitas Sebelas Maret Bidikmisi Applicant's Classification using C4.5 Algorithm

Abstract

Bidikmisi scholarship is a scholarship for poor but outstanding students. Because of the amount of applicants,
there is a need to use an accurate method in the selection process of Bidikmisi scholarship,
especially in Universitas Sebelas Maret’s (UNS) environment. In this paper, C4.5 algorithm is proposed
as a method to help on Bidikmisi recipients selection process. The dataset which is used is
Bidikmisi applicants data from 2013 to 2015. The applicant’s data from 2013 and 2014 is used as
training data and the applicant’s data from 2015 is used as testing data. Furthermore,
oversampling and undersampling technique is used to address the class imbalance problem in
training data. Finally the accuracy for each decision trees are compared to see which sampling method
is better. The result of this study shows that the accuracy of the C4.5 algorithm decision tree with
the applicant’s data from 2015 as testing data is 79,80% and Area Under Curve (AUC) value 0.5539.
Meanwhile, to compare the sampling method, the best decision tree based on testing result is chosen.
Oversampling technique produce 82,69 % for precision, 91,22 % for recall, and 77,16 % for accuracy.
While undersampling technique produce 82,78 % for precision, 91,22 % for recall, and 77,27 % for accuracy.
Therefore it is concluded that undersampling technique gives a better accuracy than oversampling technique.