Analisis Performa Random Forest dalam Deteksi Diabetes Menggunakan Fitur Manual Citra Retina dan Teknik SMOTE berbasis Website

Abstract

The importance of early diabetes detection drives the development of efficient screening technologies. This research implements and analyzes the performance of a Diabetic Retinopathy (DR) detection system using a classical machine learning approach. The system utilizes the Random Forest algorithm trained on a combination of manual features (color, shape, and texture) extracted from retinal fundus images. To address the common issue of imbalanced datasets in medical data, the Synthetic Minority Over-sampling Technique (SMOTE) was applied to the training data. The evaluation results on the test set show excellent performance, with the model achieving an accuracy of 94.82% as well as balanced precision and recall scores of 0.95 for both "Normal" and "Diseased" classes. This finding proves that with appropriate feature engineering and proper data balancing, classical methods like Random Forest can serve as a highly effective and reliable tool for preliminary DR sc lebih akurat dan andal secara klinis