Diabetic Patient Calorie and Food Management System for the Indian Diet

Type 2 diabetes mellitus has become an acute public health issue in India because of the interaction of genetic susceptibility, rapid urbanization, and culturally diverse, high-carbohydrate diets. The paper introduces a hybrid machine learning architecture that combines deep learning and personalised modelling to provide culturally adaptive diet suggestions for diabetic patients. We employ ResNet-50 for food image recognition and portion size estimation, and a Random Forest (RF) model for estimating caloric intake requirements based on patient demographics, health status, and activity levels. The system dynamically considers glycemic index values and local dietary habits to modify meal suggestions. Our experimental results on a self-generated dataset of 200 Indian foods achieved 93% accuracy in food recognition and ±7% error in caloric estimation, with mAP of 85.2% and recall rates over 90% in the majority of food categories. The paper concludes with a discussion of integration challenges and future extensions towards real-time mobile deployment and IoT integration.