A Framework for Identifying the Learners’ Engagement in E-learning Environments Using a Hybrid CNN Architecture

In the realm of online education, precise identification of the participants and their engagement is critical for optimal learning results. With the aim of precisely monitoring attendance and engagement in an online learning environment, this study suggests an end-to-end framework based on a hybrid convolutional neural network (CNN) architecture. The complete evaluation employs two separate CNN models: CNN-I for identifying learners’ faces in individual images and CNN-II for measuring engagement through facial expressions, eye gaze and head movement. Two custom datasets have been developed and annotated to allow for an in-depth assessment of the proposed approach. Dataset-I contains 4000 individual face images, whereas Dataset-II has 6720 images depicting various interactions of the learners, such as engaged, bored, frustrated, confused, drowsy and looking away. However, on Dataset-I and the VGG-face dataset, the CNN-I model surpasses a deep CNN (DCNN) with superior accuracy rates of 95.25% and 88.60%, respectively. On Dataset-II and the FER2013 dataset, the CNN-II model obtains similarly outstanding accuracy rates of 93.45% and 83.74%, respectively. The specialised CNN-I and CNN-II models outperform the hybrid CNN model in their particular tasks, albeit having an accuracy rate of 88%. The research highlights the enormous potential of straightforward architecture integrated with specific datasets, emphasising their computational effectiveness and efficiency in the domains of engagement and facial recognition for online learning.