An AI-Powered Predictive Framework for Calorie Estimation in Traditional Malay Cuisine

Abstract

Traditional Malay cuisine poses a significant challenge for automated calorie estimation due to its intricate textures, diverse ingredient compositions, and heterogeneous portioning styles. This paper proposes an end-to-end artificial intelligence-driven predictive framework that estimates the caloric content of Malay dishes directly from food images. The system integrates a fine-tuned convolutional neural network for robust dish recognition, a regression-based portion-weight estimation module leveraging deep visual features, and a nutritional computation engine grounded on standardized food composition databases. The framework is trained and validated using a multi-dataset approach comprising Food-101, UEC-Food256, and a curated Malay Food Dataset to enhance both generalization and cultural specificity. Experimental evaluation reveals high classification performance and a mean absolute percentage error of 9.12% for calorie prediction, outperforming existing image-based nutrition estimation techniques for culturally complex cuisines. The findings highlight the feasibility of deploying AI-based calorie monitoring systems tailored to regional food practices and the potential of culturally aware computational nutrition models in supporting personalized dietary assessment, public health management, and intelligent food tracking applications.