A Food Recommender System for Type 2 Diabetic Patients using Singular Value Decomposition

Abstract

Background: Diabetes mellitus affects insulin production or response and causes high blood sugar levels. Personalized health/wellness recommendations are indispensable in world and are provided by health recommender systems such as Singular Value Decomposition (SVD). Objective: To create a food recommender system for type 2 diabetic patients using SVD Methods: A questionnaire was designed with a validation process to obtain the information of the participants in the interview, following a comprehensive literature review. A total of 917 patients, at Matugama district hospital were selected using a random sampling method, and a complete dataset containing fasting blood sugar (FBS) level, age, gender, weight, height, and preferences for 50 diabetic-friendly food items (for Sri Lankan meal style) were obtained through a literature review and consultation with a nutritional expert. Data were collected using direct patient interviews and from data available in the literature. The SVD technique recommended food items based on the Cosine Similarity Metric and user-based Collaborative Filtering. Most similar patients to a particular patient were identified, and the highest-rated food items were ranked considering the glycemic index and glycemic load. The system was implemented using the Python programming language, with essential libraries such as NumPy and Pandas being leveraged and Mean Absolute Error used as validation metric. Results: Patients were assisted in achieving their blood glucose targets and the risk of unexpected blood spikes and dips was reduced. Nutritional foods with low glycemic index and glycemic load were recommended by the system based on FBS level and ratings given by patient for foods. A specific cut-off FBS level of 126 mg/dL or higher is used to tailor recommendations. The ability to handle a new patient is also incorporated and prediction accuracy of system was 89%, demonstrating its strength and reliability in providing personalized dietary suggestions.

Conclusion: The SVD systems can be implemented to recommend food items for type 2 diabetic patients, improving their adherence to dietary guidelines and managing their blood sugar levels effectively. It is recommended for further modifications promoting the commercialization of health informatics.