The rapid growth of digital food datasets and the increasing demand for automated dietary monitoring systems have created a need for efficient food classification techniques. Traditional machine learning approaches often struggle to handle the high-dimensional and complex nature of food images, mainly due to limitations in manual feature extraction and scalability. To address these challenges, this research proposes a hybrid framework that combines transfer learning models with machine learning classifiers for accurate food image classification.In the proposed approach, pre-trained deep learning architectures are utilized to extract meaningful visual features from food images, enabling the system to capture complex patterns and representations. Models such as EfficientNet, DenseNet, and MobileNet are employed as feature extractors due to their strong performance in image recognition tasks. The extracted feature vectors are then classified using machine learning algorithms including XGBoost and Random Forest to improve prediction accuracy and interpretability.The hybrid framework integrates the feature learning capability of deep neural networks with the decision-making efficiency of classical machine learning algorithms. Experimental evaluation demonstrates that this combination improves classification accuracy and robustness, even when dealing with noisy or diverse food image datasets. The results indicate that the proposed system can effectively classify multiple food categories and can be applied in real-world applications such as nutritional monitoring, automated dietary assessment, and food safety management systems.
Professor et al. (Thu,) studied this question.