Comparative Analysis of Deep Learning Models for Fashion Recommendation in E-Commerce

In an era marked by a rapidly growing population and an increasing demand for online marketing and shopping, the role of recommendation systems becomes paramount. This research focuses on leveraging deep learning techniques to enhance fashion recommendation systems in the context of e-commerce websites. The Myntra Fashion Dataset is used in this research to explore the utilization of ensemble techniques to assess the integrity and compatibility of deep learning models, specifically VGG16 and MobileNet. The study investigates the effectiveness of these models individually and in hybrid combination, focusing on their ability to enhance the accuracy and efficiency of the recommendation system. Through rigorous experimentation and evaluation, the key findings indicate that while VGG16 and MobileNet demonstrate effectiveness with highest similarity score as 94, 100 for top 5 product recommendation when used alone and outperformed rest of the models such as Inception, DenseNet, ResNet50, MobileNetV3Small, and hybrid (VGG16 + MobileNet) where their compatibility and integrity as an ensemble are not as efficient. The research underscores the importance of carefully considering model integration and highlights the need for further exploration to optimize the performance of ensemble models in recommendation systems. These findings contribute to the ongoing advancement of deep learning techniques in the e-commerce domain, supporting effective buying decisions and increasing profitability for traders.