A Multinutrient Clustering Framework for Personalized Food Recommendations

Area of Science:

• Computational Nutrition
• Nutritional Informatics
• Personalized Health

Background:

• Personalized nutrition recommendations struggle with large food databases and user preferences.
• Existing methods lack effective alignment between dietary needs and available food options.

Purpose of the Study:

• To develop and evaluate a multinutrient clustering framework for personalized food recommendations.
• To improve the accuracy and relevance of dietary suggestions by considering multiple nutritional attributes.

Main Methods:

• Analyzed 8790 foods using 23 nutritional attributes from the USDA database.
• Applied and compared K-means and agglomerative clustering algorithms, optimizing K-means with eight clusters.
• Evaluated performance using silhouette scores, Precision@K, NDCG, and fivefold cross-validation.

Main Results:

• K-means clustering with eight clusters achieved optimal performance (silhouette score 0.273), yielding interpretable dietary categories.
• The proposed framework demonstrated significant improvements: 46.4% over popularity-based and 273.0% over single-nutrient recommendations.
• The system provides real-time, personalized recommendations (under 2 seconds) based on user-defined nutrient preferences and weighting strategies.

Conclusions:

• The multinutrient clustering framework offers an effective solution for personalized nutrition recommendations.
• This research bridges computational nutrition with practical dietary advice, providing an open-source system for evidence-based meal planning.
• The system addresses limitations in current prediction and classification methods by enabling personalized, multinutrient food suggestions with transparent reasoning.