A Hybrid Approach to Fashion product Recommendation and Classification: Leveraging Transfer Learning and ANNOY

Fashion recommendation systems have become indispensable for modern e-commerce platforms, aiming to deliver personalized product suggestions that enhance user satisfaction. However, the dynamic nature of fashion trends, coupled with the vast and complex landscape of fashion images, presents significant challenges for traditional recommendation approaches. This research introduces a novel hybrid framework that effectively addresses these challenges. By integrating deep learning techniques, specifically the EfficientNetB0 model, for robust feature extraction and the Annoy (Approximate Nearest Neighbors Oh Yeah) library for efficient similarity search, the system demonstrates superior performance in generating accurate and timely fashion recommendations. This approach effectively captures the intricate patterns within fashion data, leading to improved user engagement and satisfaction. Apart from the fashion recommendation, the proposed model is also capable of multi-class classification of the products with a validation accuracy of $\mathbf{9 9. 7 9} \%$ and a validation loss of 4.8%.