Prediksi Diabetes Berdasarkan Hipertensi dan Obesitas Menggunakan Algoritma Logistic Regression di Puskesmas Setiabudi

This study aimed to apply logistic regression to predict diabetes status among patients at Setiabudi Public Health Center and to analyze the relationship between predictor variables and the predicted probability of diabetes. The secondary dataset comprised 2,013 patients: 1,962 with a Normal status and 51 with a Diabetes status, yielding a class ratio of approximately 38:1. The model used age, gender, blood pressure, body mass index (BMI), waist circumference, and random blood glucose as predictors, with blood glucose status as the target. Model performance was evaluated using 10-fold cross-validation with out-of-fold (OOF) predictions, in which each patient received a prediction while their record was in a validation fold. The model achieved an accuracy of 99.95%, precision of 98.08%, recall of 100.00%, and an F1-score of 99.03%. These results should be interpreted in light of the class imbalance. Moreover, OOF evaluation provides internal validation on the study dataset rather than validation on independent or external test data. The results therefore do not establish how well the model would generalize to other populations or datasets.