Classification of Wrist Accessories: Advanced Watches with Logistic Regression, SVM, and Deep Features from Inception V3 and VGG-19

This study proposes a hybrid classification framework for wrist accessories (Analog, Automatic, Digital, and Smartwatches) by combining deep learning-based feature extraction (Inception V3 and VGG-19) with traditional classifiers (Logistic Regression and SVM). The Inception V3 + Logistic Regression model achieved the highest performance, with 95% accuracy and an AUC of 0.999. t-SNE visualization revealed distinct clusters for Digital and Smartwatches, while Analog and Automatic categories exhibited partial overlap, indicating challenges in distinguishing visually similar classes. The findings underscore the effectiveness of deep features in improving classification accuracy. Practical applications include retail inventory management, counterfeit detection, and e-commerce product sorting, highlighting the model's real-world adaptability and scalability. This approach demonstrates the potential of hybrid techniques in enhancing automated classification systems for wrist accessories.