An Ensemble Deep Learning Framework For Student Attentiveness Detection Using The Daisee Dataset

Abstract

Student attentiveness plays a critical role in determining learning effectiveness in modern smart classroom and e-learning environments. Manual monitoring of student engagement is often subjective, inconsistent, and difficult to perform continuously in large-scale digital learning systems. To address this challenge, this paper proposes an ensemble deep learning framework for automated student attentiveness detection using the DAiSEE dataset. The proposed approach integrates a MobileNetV2-based convolutional neural network for deep facial feature extraction with classical machine learning classifiers including Support Vector Machine (SVM) and Random Forest (RF). The outputs of individual classifiers are combined through a weighted soft-voting ensemble strategy to improve classification robustness and prediction reliability across multiple attentiveness levels.

The DAiSEE benchmark dataset, containing real-world e-learning video recordings with diverse illumination conditions, facial expressions, and head poses, was used for experimental evaluation. Video clips were converted into image frames and preprocessed through normalization and resizing before feature extraction and classification. The proposed framework classifies student attentiveness into four engagement categories: Very Low, Low, High, and Very High. Experimental results demonstrate that the ensemble framework outperforms individual classifiers, achieving an overall accuracy of 91.7%, macro-average AUC of 0.915, and improved F1-score performance. Confusion matrix and ROC curve analyses further confirm the effectiveness of the proposed approach in distinguishing attentiveness levels under realistic learning conditions.

In addition, a GUI-based decision support interface was developed to visualize attentiveness predictions and assist instructors in real-time classroom analytics. The proposed framework provides an efficient, scalable, and interpretable solution for AI-driven student engagement monitoring in intelligent educational environments.