Deep learning-based model for analyzing student engagement in activities

Student engagement is a critical determinant of academic success. Traditional methods like surveys and self-reporting are subjective, while automated models often fail to integrate multimodal behavioral cues effectively, resulting in inaccurate or incomplete engagement analysis. To address these limitations, this research introduces an advanced framework, the Intelligent Chimp-driven Bi-directional long short-term memory network-scalable gated recurrent unit (IC-BiSGRU-Net), for robust analysis of student engagement in diverse learning activities. Multimodal educational dataset were collected from multimodal sources, including classroom video recordings, audio interactions, and digital activity logs. Pre-processing involves spectral median filtering for audio signals, min-max normalization for video sequences, and digital logs to handle scale, illumination, and distribution variations. For feature extraction, facial micro-expressions and action units were derived using a Convolutional neural network (CNN)-based encoder, capturing subtle emotional and attentional states. These extracted features, along with sequential interaction logs, were processed using the IC-BiSGRU-Net model, which integrates the temporal modeling power of Bi-LSTM with the computational efficiency of SGRU. Experimental validation on benchmark student engagement revealed that the proposed model outperformed the conventional model, achieving accuracy, recall, precision, and F1-score ranging from 96 to 99%. The system successfully classified engagement into active, passive, and disengaged states in real time.