Deep Learning-Based Integrated System for Student Attendance and Emotion Recognition in Elementary Education

Abstract

Mental health challenges among elementary students have become a critical global concern, with 10–20% of children experiencing emotional difficulties that significantly impact educational performance and social development. This research presents a deep learning-based integrated system that simultaneously performs automated attendance tracking and real-time emotion recognition for elementary education settings. The system employs FaceNet with one-shot learning for face recognition-based attendance, requiring only a single reference image per student. For emotion detection, we implement and compare MobileNet-v2 and YOLOv8 on an augmented FER-2013 dataset with seven emotion categories (happy, sad, angry, surprised, scared, neutral, and sleepy). Experiments conducted in an offline simulation setting show that YOLOv8 outperforms MobileNet-v2 across all metrics, achieving 65% accuracy versus 53%, with higher precision (66% vs. 53%), recall (65% vs. 55%), and F1-score (65% vs. 52%). YOLOv8’s anchor-free detection and optimized inference pipeline provide operational advantages essential for real-time classroom applications. This integrated system offers educators actionable insights for early identification and timely intervention related to students’ emotional well-being while streamlining administrative tasks. Unlike previous studies that focused on either attendance or emotion recognition separately, this research offers a novel, integrated real-time system tailored for elementary education settings.