Komparasi Perbandingan Algoritma C4.5, Naive Bayes, K-Nearest Neighbor, Random Forest Untuk Prediksi Faktor Penyebab Penyakit Diabetes

Abstract

Diabetes is a chronic metabolic disease characterized by elevated blood glucose levels and can cause various serious complications and contribute to high mortality rates worldwide. The main problem in managing diabetes is the need for accurate patient status classification based on laboratory test data so that appropriate treatment can be carried out. This study aims to compare the performance of the C4.5 algorithm, Naive Bayes, K-Nearest Neighbor (KNN), and Random Forest in classifying diabetes patient data. The dataset used was sourced from Electronic Health Records (EHRs) with research subjects from Rantauprapat Regional General Hospital, totaling 10,000 data consisting of eight attributes and one class attribute, with 859 diabetes patient data and 9,141 non-diabetes patient data. The research method was carried out by dividing the data into training data and testing data using a ratio of 90:10, 80:20, and 70:30. Evaluation of model performance used accuracy parameters and Receiver Operating Characteristic (ROC) with Area Under Curve (AUC) values. The results showed that the C4.5 and Random Forest algorithms produced higher accuracy values ​​than Naive Bayes and KNN, especially at training data ratios of 90%:10% and 70%:30%. Based on the ROC evaluation, the Random Forest algorithm obtained the highest AUC values ​​at the 70%:30% ratio of 0.972 and 80%:20% of 0.970. Based on these test results, it can be concluded that the C4.5 and Random Forest algorithms have relatively better performance and are almost equivalent in classifying diabetes based on accuracy and AUC values.