A framework for Modeling Nutritional Quality in Processed Foods via AI-Driven Salience

Thesis Abstract

This study addresses the rising complexity of nutritional quality assessment in processed foods amid expanding product reformulations and consumer demand for healthier options. It identifies a gap in integrated frameworks that couple nutritional quality metrics with AI-driven salience for prioritizing reformulation targets under real-world production constraints. The aim is to develop a scalable framework that models nutritional quality in processed foods through AI-driven salience, enablingPredictive prioritization of formulation changes that maximize healthfulness while preserving sensory and economic viability. Specific objectives are (i) to operationalize a composite Nutritional Quality Index (NQI) incorporating macronutrient balance, micronutrient adequacy, and additive safety; (ii) to design an AI-driven salience mechanism that weights NQI components by dietary guidelines, consumer preferences, and regulatory considerations; (iii) to embed the salience model within a probabilistic graphical framework linking ingredient-level decisions to product-level nutritional outcomes; (iv) to validate the framework using empirical data from a multi-category processed food dataset; and (v) to evaluate trade-offs between nutritional improvements and sensory acceptance via scenario analysis. The study employs a mixed-methods, explanatory sequential design. The population consists of commercial processed foods across three categories cereals-based snacks, ready-to-eat meals, and dairy-alternative beverages, drawn from five major manufacturers. A purposive sample of 300 SKUs is collected, with 60 SKUs per category, ensuring diversity in ingredient profiles, fortification strategies, and pricing. Data collection instruments include (a) nutrition panels and ingredient lists from product labels; (b) laboratory analyses for macronutrient and micronutrient content on a representative subset of 120 products; (c) sensory and consumer preference data from a structured consumer panel of 500 participants using a 9-point hedonic scale; and (d) regulatory and dietary guidelines metadata from recognized authorities. The AI-driven salience component integrates feature extraction from ingredient-level data, historical reformulation records, and nutrient-density signals. Analytical techniques comprise regression-based modeling to construct the NQI, Bayesian networks to model causal relations among ingredients, nutrients, and sensory outcomes, and machine learning methods (random forest and gradient boosting) to derive salience scores for reformulation targets. Validity and reliability are established through cross-validation, test-retest procedures, and triangulation with expert nutritionists. The conceptual model is implemented in a modular framework comprising data ingestion, NQI computation, salience scoring, and optimization-ready outputs for formulation decision-making. Key findings are anticipated to demonstrate that AI-driven salience can identify high-impact reformulation targets that enhance overall NQI by enabling targeted reductions in added sugars and saturated fats while maintaining or improving palatability and cost-competitiveness. Preliminary simulations are expected to reveal that a 15–25% substitution of specific additives with nutrient-dense alternatives yields statistically significant improvements in NQI (p < 0.05) without perceptible degradation in sensory scores (?1.0 ± 0.8 on the hedonic scale). The Bayesian network is expected to reveal robust conditional dependencies between certain fortified ingredients and micronutrient adequacy, offering actionable insights for shelf-stable product design. The study also anticipates identifying category-specific salience profiles, indicating that cereals-based snacks benefit most from fiber fortification, while dairy-alternative beverages benefit from micronutrient fortification aligned with target consumer demographics. The study contributes to knowledge by presenting a novel integrative framework that fuses nutritional science with AI-driven salience to guide evidence-based reformulation in processed foods. It advances theoretical understanding of how composite nutritional indices interact with perceptual and economic constraints under real-world production settings. Practically, it offers food manufacturers a decision-support tool that prioritizes reformulation actions, reduces trial-and-error iterations, and aligns product portfolios with dietary guidelines and consumer expectations. The main conclusion is that AI-driven salience is a viable mechanism to optimize nutritional quality in processed foods without sacrificing market performance, provided that the framework is calibrated with category-specific constraints and validated across diverse product configurations. Recommendations include embedding the framework in pilot-scale reformulation projects, expanding the dataset to include international products for cross-cultural validation, and integrating lifecycle assessment to quantify environmental implications of nutrition-driven reformulations.