Abstract
The convergence of artificial intelligence (AI) and nutritional science is reshaping how dietary intake is measured, how malnutrition is detected, and how personalized dietary guidance is delivered. This review synthesizes recent advances across three interconnected domains: (1) AI-based mobile dietary assessment, (2) AI for childhood nutritional classification, and (3) AI-driven individualized food recommendations. We identified and analyzed over 290 papers retrieved from SciSpace, PubMed, Google Scholar, and ArXiv, focusing on peer-reviewed studies published primarily between 2018 and 2026. Mobile dietary assessment systems employing convolutional neural networks (CNNs) and 3D volumetric reconstruction have achieved accuracy comparable to trained dietitians in controlled settings, though field validations in low- and middle-income countries (LMICs) reveal persistent challenges in portion estimation and cross-cultural generalizability. In childhood nutrition, multimodal fusion of anthropometric imaging and socioeconomic tabular data using ensemble methods (XGBoost, Random Forest) and deep learning has yielded area-under-the-curve (AUC) values up to 0.95 for malnutrition screening, with promising deployment potential on edge devices. Personalized food recommendation systems have evolved from collaborative filtering to large language model (LLM)-powered frameworks, with federated learning architectures enabling privacy-preserving, culturally adaptive meal planning. Cross-cutting challenges include data heterogeneity, model explainability, equity across populations, and the absence of large-scale randomized clinical trials linking AI outputs to health outcomes. We conclude with a roadmap for future research priorities including multimodal dataset development, clinical validation, and responsible AI deployment in nutrition care.