Development of an Image Processing based Student Engagement Prediction in Online Classes by using Deep Learning Strategy

Online education has rapidly evolved, and with it, the need for effective monitoring of student engagement. This study presents an innovative approach for predicting student engagement in online classes using a deep learning strategy that combines Inception modules and an Artificial Neural Network (ANN). The system processes real-time video streams and static images to identify and classify student engagement levels, including attentive, distracted, and confused states. A comprehensive dataset of 10,000 samples was collected from online class recordings, carefully annotated and augmented through techniques such as rotation, scaling, and brightness adjustment. The pre-trained Inception model was fine-tuned on this dataset to extract high-level discriminative features, which were then fed into a multi-layer ANN for classification. The integrated model achieved an overall accuracy of 92.5%, with precision and recall values of 90.2% and 93.0%, respectively. Detailed experiments reveal that the model performs consistently across varied classroom scenarios, offering reliable real-time predictions. Performance metrics such as F 1 score (91.6) and ROC-AUC (0.95) further substantiate the system's effectiveness. The system's design not only ensures robust performance but also adheres to ethical guidelines regarding data collection and privacy. By providing educators with actionable insights, this research paves the way for responsive teaching strategies and improved student monitoring in online environments. The promising accuracy values and the efficient processing time of 25 milliseconds per image underline the practical potential of the proposed system. Future developments will explore multi-modal data integration and advanced temporal analysis to further enhance engagement prediction.