Multi-Task Deep Learning with SHAP Explainability for Personalized Nutrition Prediction

Abstract

The purpose of this article is to address key gaps in the current personalized nutrition recommendation models. These gaps include limited personalization, limited explainability, and single-nutrient assessment/prediction. This study develops a multi-task deep neural network machine learning model to predict multiple dietary components simultaneously by taking into account individual genetic, phenotypic, and lifestyle factors. The study uses publicly available datasets that are sourced, pre-processed, and partitioned into training and test sets. Data pre-processing steps ensure data quality. Model performance is assessed using RMSE, MAE, and the coefficient of determination (R²). Model interpretability is enhanced through SHAP-based explanation techniques, which transparently elucidate feature contributions to model predictions. The proposed model offers comprehensive, personalized, and interpretable nutrition recommendations, with the goal to improve user trust, adoption, and dietary decision-making. This study contributes scalable, evidence-based methodologies advancing personalized nutrition through multi-nutrient prediction and explainable AI.