Comparison of Naive Bayes and K-NN Method on Tuition Fee Payment Overdue Prediction

Abstract

Attending basic education is an obligation for all Indonesian citizens. The financial cost is one of the input components to implement an education or even can be considered as the main requirement in achieving the goal of education. For a private education institution in Indonesia, the financial cost is mainly covered by students’ tuition payments. SMK Al-Islam Surakarta is a private school that manages all its students to pay school tuition fees monthly. According to its last year’s administrative report, the number of students who are late in paying school tuition fee is around 60%. Since the school’s operational costs are heavily depended on their income from tuition fees, this considered an essential problem and has to be managed and predicted as well. This research will discuss techniques in predicting the late payment of tuition fees. From many popular methods available in this area, we observed two of them namely Naive Bayes and K-Nearest Neighbor (K-NN). This study will compare the accuracy between those two methods. The data used for the lab work is the official education basic data of Al-Islam Surakarta Vocational School in 2017/2018 totaling 236 data. To increase its accuracy, this study also combines the prediction methods with feature selection technique Information Gain which is commonly used to select an optimal parameter for the prediction process. In the end, the system is tested using the Confusion Matrix method. The results showed that the Naive Bayes Method with Information Gain attribute selection produced the highest accuracy of 69%.