A Neural Network Method for Assessing the Nutritional Requirements of a Patient With Type 2 Diabetes

Diet change is a crucial lifestyle management recommendation for diabetic patients as it affects plasma glucose concentrations. However, the optimal macronutrient composition for plasma glucose control in Type 2 diabetes remains unclear, with researchers still using generic diabetes guidelines. This study developed a neural network classifier to determine the nutritional requirements for four types of Type 2 diabetic patients: those with chronic hyperglycemia, hypertension, obesity, or all three conditions. The classifier was trained using Kenya Food Composition Tables data on food nutrients and processing techniques. The neural network had five layers, including three hidden layers with 10 neurons each, using tanh and sigmoid activation functions, gradient descent optimization, cross‐entropy loss function, and 0.1 learning rate. Training used 40 hidden neurons per layer, 60,000 epochs and 0.2 learning rate. The neural network was evaluated against random forest, decision tree, and logistic regression models using accuracy, precision, recall, F1‐score, and Matthews correlation coefficient (MCC). The neural network achieved high performance with 91.4% accuracy, 88% recall, 86.8% precision, and 87.3% F1‐score. For the imbalanced dataset, the MCC score for the neural network was 0.808, indicating promising results for diabetes nutritional management.