Personalized Dietary Approaches for Chronic Disease Management

This project develops a preference-aware inverse optimization + clustering framework for precision nutrition, aimed at producing diet recommendations that jointly reflect (i) expert dietary constraints (e.g., DASH-style nutrient limits) and (ii) latent, segment-specific dietary preferences that differ across patient subgroups. The method unifies unsupervised clustering (to capture non-homogeneous populations) with inverse optimization (to recover utility functions and generate optimal, feasible recommendations), and it is evaluated using NHANES daily food-intake data. The approach is designed to improve guideline adherence while producing cluster-level "representative" diets that remain aligned with observed preference patterns, including handling informative but infeasible observations under the imposed dietary constraints.