An Explainable Hybrid Learning Model for Indian Food Image Classification

As the prevalence of diet management applications continues to rise for personal health tracking, the need for accurate food image classification becomes paramount. While various food classification models exist, understanding and categorizing Indian cuisine remain challenging. Although, deep learning models provide state of the art performance, they lack interpretability in terms of identifying the significant features that lead to accurate classification. In this paper, we propose an Explainable AI backed classification model specifically tailored for Indian food images. We explore the efficacy of multiple deep features learned through standard CNN architectures in conjunction with machine learning algorithms for Indian food classification. In this regard we observe that the features learned from InceptionNet and MobileNet provided superior performance of around 89% with SVM as the classifier. Further we leverage Explainable AI frameworks such as SHAP and LIME to provide insights into the model’s decision making process. Through SHAP summary plots we provide understanding of the important deep features contributing for enhanced classification and LIME provides heatmaps that highlights the significant patches in the input image which contributed to the decision.