Machine Learning-Powered Personalized Food Recommendation System for Diabetic Patients

The International Diabetes Federation (IDF) predicts that the number of people affected by diabetes will rise to 552 million by 2030, underscoring the growing global burden of this habitual condition. In this paper, we propose a machine learning-powered personalized food recommendation system designed for diabetic patients. The system combines user food preferences with nutritional needs to deliver personalized dietary recommendations that promote effective blood sugar control. The methodology employs Fuzzy-Wuzzy and Cosine Similarity as similarity analysis methods, combined with multiple machine learning algorithms, including Decision Tree, K-Nearest Neighbours, Support Vector Machine, Logistic Regression, and Naive Bayes. The system was developed using a dataset that contains 16,832 food samples with nutritional information. After preprocessing and refinement, a total of $\mathbf{1, 0 0 0}$ valid samples were retained for subsequent analysis. Performance evaluation revealed that ensemble methods significantly enhanced model accuracy, with Decision Tree (Bagging) achieving the highest accuracy of $\mathbf{8 2. 5 \%}$ and an F1-score of $\mathbf{0. 8 2 6}$. This study offers practical benefits to diabetic patients by providing them with tailored food recommendations that cater to both their nutritional needs and personal preferences.