A Novel Student Engagement Level Detection and Emotion Analysis Using Ensemble Learning

Student engagement is crucial for effective learning and monitoring, enhancing the teaching and learning experience, especially in higher education. This work proposes a mechanism for detecting student engagement levels and analyzing their emotions using computer vision and ensemble learning. The approach combines emotion detection and engagement classification models based on VGG16, ResNet50, DenseNet121, and a custom CNN trained on a dataset of student facial expressions. A head pose estimation module assesses head orientation and gaze direction, integrating these outputs with the emotion and engagement models. This holistic approach quantifies engagement by considering emotional states and head pose. Additionally, Gradient-weighted Class Activation Mapping (Grad-CAM) is used to provide visual explanations for the model’s predictions. The method is evaluated on a dataset of student images, demonstrating its effectiveness in accurately detecting engagement levels and emotional states.