SmartFace: An Offline-First AI-Based Attendance Management System for Rural Schools Using MobileFaceNet 29 23

Abstract

Manual recording of student attendance in rural schools is a time-consuming and error-prone task. Moreover, attendance records are necessary for keeping a check on educational activities and promoting social welfare schemes. In literature, most existing attendance automation systems rely on continuous internet connectivity, cloud-based processing, or costly hardware infrastructure, which can limit their scalability, affordability, and ease of deployment. Hence, a lightweight attendance system named SmartFace has been designed and developed in this research work which uses facial recognition technology for marking students' attendance digitally. SmartFace makes use of MobileFaceNet, a Convolutional Neural Network model that allows facial recognition on the device itself, conserving processing resources and eliminating the need for server-side processing. The proposed system is developed as a Progressive Web Application (PWA) and it can function without an internet connection and synchronize data with Firebase Firestore once connectivity is restored. The MobileFaceNet model of SmartFace application was trained and tested on the Indian Faces Image Classification Dataset obtained from Kaggle data repository. On evaluation, the fine-tuned model achieved a training accuracy of 99.43%, with a validation accu-racy of 93.75% and a test accuracy of 96.88%, demonstrating the viability of the proposed approach on the dataset used. These findings suggest the potential of lightweight facial recognition technology as a viable approach for ef-ficient and economical attendance management in rural schools, warranting further evaluation on larger and more diverse datasets.