Title: A Machine Learning Framework for Personalized Nutrition and Dietary Adherence Prediction in MASLD Patients
Authors: Thao Tran Vy, Quang-Truong Tran, Lan Hoang Thi
Volume: 10
Issue: 5
Pages: 54-63
Publication Date: 2026/05/28
Abstract:
Metabolic Dysfunction-Associated Steatotic Liver Disease (MASLD) requires sustained lifestyle change, yet real-world outcomes are constrained by variable dietary adherence. Most AI nutrition systems estimate energy and macronutrient targets but ignore a patient's likelihood to follow the prescription. This proof-of-concept study evaluates whether machine learning (ML) can (i) predict dietary adherence from routine clinical biomarkers and (ii) integrate that prediction into a personalized nutrition recommender for MASLD. A synthetic cohort of 2,000 patient-like records was generated to reflect plausible MASLD physiology. Inputs included BMI, body-fat percentage, fasting glucose, total cholesterol, and liver fat index (CAP), with age/weight/height available for derived metrics. Random Forest and Gradient Boosting were used for multi-target regression of calorie, protein, fat, and carbohydrate needs; Logistic Regression classified adherence (High vs. Low). On a held-out test set, adherence prediction achieved 85.2% accuracy and AUC ? 0.87 while macronutrient/energy prediction reached Rē = 0.789-0.847. The best models were deployed in a microservice architecture (FastAPI backend, Next.js frontend) to produce individualized 7-day meal plans with real-time feedback. Although generalizability is limited by synthetic data and the absence of clinical validation, results support the feasibility of ML-assisted adherence prediction and its utility for behavior-aware, precision nutrition in MASLD. Future work will validate with real cohorts, report confidence intervals and statistical tests versus clinical/statistical baselines, and add SHAP-based interpretability for clinician-facing explanations.