AI-Driven Prediction of Food Nutrient Composition and Personalized Dietary Recommendation: Advances in Multimodal Methods, Knowledge-Enhanced Decision-Making, and Evaluation Frameworks

Reliable personalized dietary advice depends on linking food identification, portion estimation, nutrient composition data, and safety rules, yet these components are often evaluated separately. This review critically examines how spectral measurements, images, text, and food databases support nutrient estimation and how uncertainty in those estimates affects dietary recommendations. Evidence on data resources, nutrient prediction, multimodal fusion, and recommendation systems is considered in light of study design, from conceptual descriptions and simulations to internal testing, held-out evaluation, validation across settings, and prospective evaluation. Reported performance depends on the experimental unit, the origin of reference values, metric definitions, and the independence of evaluation data. Spectral models may perform differently across instruments and food matrices; image segmentation does not establish portion mass or account for hidden ingredients; and values generated from text require verification against identifiable composition sources. These differences limit direct comparisons of accuracy across studies. Across the recommendation pathway, reliable decisions require traceable data and rules, uncertainty estimates calibrated on independent data, and checks of dietary constraints before ranking by preference. Nutrient limits must be interpreted over the period specified by their source, while incomplete or conflicting clinical information calls for professional review. Current evidence supports selected components more strongly than the full pathway. External validation across devices, cuisines, and populations remains limited, and prospective studies are needed to examine actual intake, adherence, constraint violations, and health outcomes. The combined performance and clinical safety of an integrated system remain unestablished.