Benchmarking Machine Learning and Transfer Learning Approaches for Fruit Classification Using Explainable Artificial Intelligence

In this study, we investigate traditional machine learning methods with respect to state-of-the-art deep learning approaches for fruit image classification. Conventional classifiers such as Random Forest, and eXreme Gradient Boosting were evaluated alongside neural networks and state-of-the-art transfer learning models, including Residual Network (ResNet), Visual Geometry Group 16 (VGG16), and EfficientNet-B0. The fine-tuned ResNet50 model obtained the maximum classification accuracy as 97.57%, substantially surpassing other traditional solutions as well as shallow networks. One important contribution of this work is in the focus on model explainability. To understand the reasoning from the deep models, we used Gradient-weighted Class Activation Mapping (Grad-CAM) and Grad-CAM++ techniques to visualize what kind of decision is helping the model making predictions as well as by considering what regions in input images are being looked up for the concerned class. This explainability integration provides the transparency and explainability of classification system, being crucial for its real world applicability. The results underline the significance of both model capability and explainability when it comes to model selection, and show that fine-tuned deep networks combined with explainability tools is a strong and useful framework for image based classification.