Yuliana Melita Pranoto, Anik Nur Handayani, Yosi Kristian
Online business has rapidly developed in the past 3 years since the COVID-19 pandemic. In Indonesia, online businesses have significantly improved, acquiring 50% more customers during this period. This study implements a specific approach to image base similar marketplace product grouping. Our intelligent system is targeted to improve users’ quality of similar product recommendations. This chapter presents an approach that utilizes artificial neural network models and deep learning techniques. A total of 4634 product images from a marketplace were used in this study. The images were categorized into 281 group labels. We utilized transfer learning from the pre-trained model by fine-tuning, adding our custom layers, and then retraining the network. We also compared VGG-16, MobileNetV2, and EfficientNetV2M. The best results were obtained from EfficientNetV2M, with an accuracy of 84%. The VGG-16 model had an accuracy of 70%. The lighter and faster MobileNetV2 model produced an accuracy of just 57%. © 2024 selection and editorial matter, Aji Prasetya Wibawa.
Universitas Negeri Malang, Malang, Indonesia; Institut Sains dan Teknologi Terpadu Surabaya, Surabaya, Indonesia