Development of a Face Detection System Using Convolutional Neural Network on an E-Commerce Website

Abstract—The rapid expansion of the e-commerce sector in Indonesia has increased the risk of digital fraud, particularly involving fake identities during seller registration. This study purposes a face detection system using Convolutional Neural Network (CNN) implemented with TensorFlow. Data were collected using a webcam, annotated via LabelMe, and augmented with Albumentations. The model was built using the VGG16 architecture in a multi-output CNN for face Classification and bounding box localization. A total of 90 high resolution images were used, divided into training, validation, and testing sets. Following 40 training epochs, the model achieved an accuracy of 91%, precision of 89%, recall of 98%, and F1-score of 93%. The model was deployed in .h5 format. Future work includes integrating Optical Character Recognition (OCR) for ID validation and Wazuh for real-time monitoring.