Scholarlink

<p>Assessing student engagement in online learning sites is crucial for effectively analyzing student involvement in the course. Given the slow and subjective nature of conventional and self-reporting methods, it has become challenging to understand student engagement. The proposed design of a working model will predict student engagement levels using a novel deep-learning framework. The system employs an ensemble model that takes spatial and temporal features extracted from video clips, with ResNet-18 for spatial feature extraction and ViViT for temporal feature processing. The features are passed through a Multi-layer Perceptron (MLP) for spatial data and Long Short-Term Memory (LSTM) for temporal data. Trained on the new real-time generated dataset, GITAM Online Learning Dataset (GOLD), the model was embedded into a real-time application capable of processing webcam input and predicting engagement every 10&#xa0;s by averaging results over 5-min windows for deeper insights. Experimental results showed that the ensemble model enhances classification accuracy compared to its counterparts, detecting engagement in real-time. This approach provides a scalable and objective student engagement monitoring solution wherein faculty can identify disengagement early and intervene to improve learning outcomes while circumventing a drawback of traditional methods and giving room for further research on automated engagement detection.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Predicting Real-Time Student Engagement in Online Learning Environment Using an Ensemble Model and the GOLD Dataset

  • 期刊论文

  • 发表日期：

    2026年4月4日

查看全文
  • Sneha Sureddy,
  • I. Jeena Jacob

摘要

Assessing student engagement in online learning sites is crucial for effectively analyzing student involvement in the course. Given the slow and subjective nature of conventional and self-reporting methods, it has become challenging to understand student engagement. The proposed design of a working model will predict student engagement levels using a novel deep-learning framework. The system employs an ensemble model that takes spatial and temporal features extracted from video clips, with ResNet-18 for spatial feature extraction and ViViT for temporal feature processing. The features are passed through a Multi-layer Perceptron (MLP) for spatial data and Long Short-Term Memory (LSTM) for temporal data. Trained on the new real-time generated dataset, GITAM Online Learning Dataset (GOLD), the model was embedded into a real-time application capable of processing webcam input and predicting engagement every 10 s by averaging results over 5-min windows for deeper insights. Experimental results showed that the ensemble model enhances classification accuracy compared to its counterparts, detecting engagement in real-time. This approach provides a scalable and objective student engagement monitoring solution wherein faculty can identify disengagement early and intervene to improve learning outcomes while circumventing a drawback of traditional methods and giving room for further research on automated engagement detection.