Diabetes Risk Classification

Screening classifier built on a 253,000-record US population health dataset. Corrected an 84% negative class imbalance with cost-sensitive weighting and precision–recall threshold optimisation to produce a model usable in a real screening context. Identified a statistically significant reversal in the BMI–sex relationship across diabetes groups (men carry higher BMI when healthy, women when diabetic), supporting sex-specific screening thresholds. Quantified that lower-income smokers face double the health risk of higher-income smokers, and translated the findings into a ranked clinician intervention framework. Assessed the disproportionate burden false-positive results place on lower-income patients as part of the analytical design.