Evaluation of Fund Usage and KJMU Potential Recipients Modeling using Classification Tree and EasyEnsemble

Abstract

(Kartu Jakarta Mahasiswa Unggul (KJMU) was one of Jakarta Provincial Government main program in education. This program aims to help students from poor family with excellent grades to continue higher education. This research was started by conducting survey to 354 recipient who entered college between 2016-2018 to see how they use KJMU funds. Even though they claim knowing how to manage their funds surprisingly education expenses come in fifth position out of 7 types of expenses and their incomes only covers 64.55% of their primary expenses. The main cause of these problems suspected because the candidates for KJMU recipients did not match the requirement. That is why this study continues by finding the right methods for classify the candidates. Since the recipient of KJMU is minority compares to majority people in Jakarta which born in 1997-2000 there is class imbalance issue in making classification model. If this issue not resolved well it will cause accuracy paradox where the prediction will tend toward majority class. This study compared CART (Classification Tree) and EasyEnsemble to find the most suitable model. Classification tree is known with its easy interpretation, high accuracy and fast but this method requires the balanced class that is why we add undersampling techniques into it. EasyEnsemble was designed for handling imbalanced class and it was combinations of UnderBagging and ADABoost. The results show that EasyEnsemble is the best method with the highest F1 score over 10, 50 dan 100 iteration modelling.