Abstract:In the realm of education, feedback emerges as a pivotal component, serving to foster engagement and interaction while also facilitating the refinement of teaching methods to capture and maintain student attention. Traditional classroom assessment methods often struggle to accurately gauge the degree of comprehension among students during lectures, relying on manual comment collection that inherently carries the risk of inaccuracies. In response to this challenge, a novel system has been proposed, harnessing the power of Facial Emotion Recognition (FER) technology to capture student feedback. Within this framework, students are given a unique avenue to convey their emotions and reactions, employing facial expressions and gestures as the means to communicate. This innovative approach enables the analysis of students' emotional responses and thereby provides invaluable insights into their comprehension levels, as well as the overall quality and engagement experienced during lectures. The approach takes shape through the utilization of Computer Vision techniques, with a particular focus on an unobtrusive methodology for assessing students' overall engagement. Overcoming limitations of traditional assessment, our approach integrates compound scaling, employing the proposed Multitask EfficientNetB0 model recognized for its proved accuracy in emotion recognition (95.7%) and behavior analysis (96.3%) across diverse datasets (DAiSEE, iSED, iSAFFE). The behavioral classification system categorizes students into "Engaged" and "Disengaged" classes within a multi-class framework, providing nuanced insights into comprehension and Student engagement. Assessment metrics, including ROC Curves, Precision, Recall, and F1-Score, ensure a thorough evaluation. Our system's adaptability is demonstrated across varied educational environments, showcasing real-world efficacy in classrooms, laboratories, and seminar halls. The inclusion of MTCNN enhances face detection capabilities, facilitating robust analysis in dynamic scenarios. Expanding its applicability, the model has been put to the test in a range of educational settings, including classrooms, laboratory environments, and seminar halls, offering dual-capability analysis of both emotions and behavior. This comprehensive approach yields nuanced insights into student engagement and interaction, and its performance has been validated through real-world deployment within classrooms and seminars

Multimodal Group Activity Dataset for Classroom Engagement Level Prediction

Multimodal Engagement Analysis from Facial Videos in the Classroom

Prediction and Localization of Student Engagement in the Wild

Unobtrusive and Multimodal Approach for Behavioral Engagement Detection of Students

Multimodal Engagement Recognition From Image Traits Using Deep Learning Techniques

CMOSE: Comprehensive Multi-Modality Online Student Engagement Dataset with High-Quality Labels

A dataset for assessing real-time attention levels of the students during online classes

Long-Term Human Participation Assessment in Collaborative Learning Environments Using Dynamic Scene Analysis

Multitask EfficientNet affective computing for student engagement detection

Classroom Interaction Behavior Recognition Dataset

MUTLA: A Large-Scale Dataset for Multimodal Teaching and Learning Analytics

Toward Automated Classroom Observation: Multimodal Machine Learning to Estimate CLASS Positive Climate and Negative Climate

Bimodal Learning Engagement Recognition from Videos in the Classroom

Real‐world multimodal lifelog dataset for human behavior study

Automated Assessment of Encouragement and Warmth in Classrooms Leveraging Multimodal Emotional Features and ChatGPT

VLEngagement: A Dataset of Scientific Video Lectures for Evaluating Population-based Engagement

Towards Student Actions in Classroom Scenes: New Dataset and Baseline

An English video teaching classroom attention evaluation model incorporating multimodal information

Detecting Behavioral Engagement of Students in the Wild Based on Contextual and Visual Data

Do I Have Your Attention: A Large Scale Engagement Prediction Dataset and Baselines