Comparison of the K-Nearest Neighbor (K-NN) and C4.5 Algorithms for Predicting the Graduation Status of Indonesia Smart Card (KIP) Scholarship Recipients at AMIK Medicom Medan

Abstract

The increasing number of students receiving the Smart Indonesia Card (KIP) Lecture requires universities to develop an early prediction system in monitoring student graduation status. This study aims to compare the performance of the K-Nearest Neighbor (K-NN) and C4.5 algorithms in predicting the graduation status of KIP Lecture recipients at AMIK Medicom Medan. The study used a comparative descriptive approach with a dataset of 310 students, where the Achievement Index (IP) from semester 1 to semester 4 was used as a predictor variable, while graduation status (Graduated or Drop Out) was used as the target class. The research stage includes data preprocessing consisting of data cleaning, coding, normalization, and data sharing. The K-NN model is optimized using 10-fold cross-validation to determine the best K-value, while the C4.5 model is built using a decision tree based on entropy criteria. Model performance was evaluated using confusion matrix, accuracy, precision, recall, and F1-score. The results showed that the K-NN algorithm with a value of K = 3 produced the best classification performance with an accuracy of 99.03%, a macro F1-score of 0.96, and a weighted F1-score of 0.99. Meanwhile, the C4.5 algorithm also showed excellent performance with an accuracy of 98.71%, a macro value of F1-score of 0.94, and a weighted F1-score of 0.99. Although K-NN results in slightly higher prediction performance, the C4.5 algorithm has an advantage in terms of interpretability through an explicit decision tree structure making it more suitable to support academic decision-making. These findings show that both algorithms are highly effective in predicting student graduation status, with K-NN excelling in terms of prediction accuracy, while C4.5 is superior in providing transparency for decision support systems in the education sector.