Diabetes Screening

SVM-RBF was preferred for screening because it combined competitive ROC-AUC with stronger recall than higher-accuracy alternatives. On the held-out test set, SVM-RBF was tuned toward sensitivity rather than precision. Successful national lifestyle intervention programmes have been implemented in the United States, Finland, Japan, India, and China, providing models for adaptation. Fasting insulin is rarely ordered by default but is highly informative because elevated levels indicate insulin resistance years before glucose rises into the diabetic range. Hemoglobin A1c reflects average blood glucose over the preceding 2–3 months and is available through standard lab work. Random Forest had the highest raw accuracy but was less suitable for screening because its recall was low. Stage 1 used the NCSU Diabetes Dataset with 13 clinical and symptom features and diabetic status as the binary outcome. The framework treats binary diabetes detection as a supervised screening task using clinical tabular data. Expanding HbA1c-based screening beyond current clinical practice is identified as an urgent need given that 60% of T2D cases were undiagnosed. BMI and waist circumference are identified a…