Classroom Activity Classification with Deep Learning

Understanding students' classroom activities has become increasingly important in educational research. Accurate classification of classroom activities can provide valuable insights into student engagement, learning dynamics, and classroom management. Traditional methods of observation and classification are often time-consuming and subjective. Therefore, this study explores the application of ML and TL algorithms for the automated classification of classroom activities, aiming to address these challenges. Moreover, we created a diverse dataset of 1,698 images sourced from platforms, including Google and educational resources. The dataset was meticulously labeled into the three primary target classes: reading, writing, and side-talking. Various ML models, such as LR, RF, SVM, and XGB, were employed to attain accurate classification. Additionally, DL algorithms, such as CNN, ResNet50, MobileNet, and Inception V3, were utilized to harness the power of transfer learning for this challenging task. The performance of these classifiers was rigorously evaluated using traditional metrics. RF attained the highest accuracy score of 89.13% among ML models, while MobileNet outperformed other TL models with an accuracy score of 99.71%. These findings provide valuable insights into the potential applications of automated classroom activity monitoring in educational research and offer a promising avenue for further investigation in this domain.