An Image-Based Recommendation System Using Voting-Assisted Ensemble Machine Learning Approach

In recent days, e-commerce platforms have grown at an unexpected level worldwide. These sites usually rely on recommendation algorithms, which employ user terms to identify related goods or products. Most e-commerce applications are based on customer reviews, which are knowledge-based recommendations. Moreover, customers wish for an interactive platform to buy their needs, where interactions increase interest. In general, recommendations based on images that solve the interaction problem are rarely studied in the literature. Hence, the proposed paper focuses on an Image-Based Recommendation System using Deep Feature Extraction and Ensemble Machine Learning Approach. In the technique, initially, the image is pre-processed by removing the background using Adaptive Otsu thresholding-based segmentation. Secondly, the Stacked ResNET pyramid dilated convolution (PDC-ResNET) model is employed to extract the high-level features from the images. The extracted features are then selected using Binary Hunger Games search optimization (BHGO). Lastly, the product’s class or kind is categorized using a voting-based Ensemble of AdaBoost XG-Gradient Boosting Decision Tree (AX-BDT). There are two main levels in the suggested recommendation system: Level 1 and Level 2. The suggested model determines the product’s class and category at Level 1. The suggested recommendation algorithm in Level 2 retrieves products that are closely matched and similar. Two datasets, the Fashion product dataset and the Amazon product dataset, are used to assess the suggested approach. In the fashion product dataset, the proposed model results in 96.99% accuracy. In the Amazon product dataset, the proposed model results in 97.94% accuracy.