Prediksi Klasifikasi Diabetes Berbasis Faktor Risiko Kesehatan Menggunakan Gradient Boosted Trees

Abstract

Diabetes mellitus is a chronic disease with a high prevalence that requires early‑stage risk detection to enable effective prevention efforts. This study aims to analyze the capability of the Gradient Boosted Trees algorithm to classify early‑stage diabetes risk based on clinical symptoms using the Early Stage Diabetes Risk Prediction dataset. The research methodology includes data preprocessing, splitting the data into training and test sets, and training a Gradient Boosted Trees classification model in RapidMiner with the class attribute set as the labeled target. Model performance is evaluated using accuracy, weighted mean precision, and weighted mean recall metrics to assess the balanced classification ability for each class. Experimental results show that the Gradient Boosted Trees model achieves good classification performance with an accuracy of 91.76%, a weighted mean precision of 92.04%, and a weighted mean recall of 90.49% on the test data, supported by a confusion matrix pattern dominated by correct predictions for both classes. These findings indicate that the Gradient Boosted Trees approach has strong potential to be used as a decision‑support component in early diabetes risk detection systems and is worth further development for broader clinical data scenarios.