Prediksi Diabetes Melitus Menggunakan Algoritma J48 pada Dataset Pima Indians Diabetes

Abstract

Diabetes Mellitus (DM) is a non-communicable, chronic degenerative disease with a significantly rising global prevalence, posing a comprehensive threat to public health. Indonesia currently ranks fifth highest in the world, with the number of sufferers reaching 19.5 million individuals. Because conventional diagnostic processes through laboratory examinations require a relatively long time, a preventive management approach based on intelligent computing utilizing data mining techniques is required. This study applies the J48 Algorithm (Decision Tree C4.5) using WEKA software to predict the risk of Diabetes Mellitus at an early stage. Model performance evaluation was conducted using the 10-fold cross-validation method on 768 data samples from the Pima Indians Diabetes Dataset, which consists of 8 maternal clinical attributes and 1 target variable (label). Prior to modeling, the dataset underwent a series of comprehensive pre-processing steps, including data cleaning, data integration, data transformation, and data reduction, to handle hidden missing value anomalies in the form of illogical zero values. The test results demonstrate that the J48 Algorithm achieved an accuracy of 73.82%, placing the Glucose attribute as the root node, making it the primary determining factor in classification decision-making. Although the model exhibits practical characteristics with a computation time of 0.01 seconds, the confusion matrix analysis reveals that 108 false-negative cases were still found. This finding indicates that the baseline model with standard parameters still requires further optimization through data-balancing techniques to enhance diagnostic sensitivity in the positive class.