ANALISIS KOMPARATIF ALGORITMA NAIVE BAYES DAN DECISION TREE UNTUK PREDIKSI PENYAKIT DIABETES BERBASIS WEB

Abstract

The number of people living with diabetes keeps rising worldwide, and many cases still go unnoticed until complications emerge. This study compared two classification algorithms, Naive Bayes and Decision Tree, for predicting a patient's diabetes status from five clinical measurements: age, glucose level, blood pressure, insulin level, and body mass index. Records were drawn from the Pima Indians Diabetes Dataset (768 patients) and split into training and testing subsets at an 80:20 ratio while preserving class proportion through stratified sampling. Prior to modeling, the data went through missing-value handling and feature scaling, after which both algorithms were tuned through Stratified 5-Fold Cross Validation combined with GridSearchCV. Each model's performance was examined through its confusion matrix, focusing on accuracy, sensitivity, and specificity. After tuning, the Decision Tree clearly surpasses Naive Bayes, reaching 87.01% accuracy, 80.40% sensitivity, and 88% specificity, while Naive Bayes only reaches 71.43% accuracy, 60.28% sensitivity, and 80% specificity. Given its stronger detection ability, the tuned Decision Tree is embedded into a web-based application that lets health workers run a quick diabetes risk screening. The resulting output serves purely as supporting information and is not meant to substitute a physician's clinical judgment.