Learning Behavior Analysis in the Classroom Based on Deep Learning

This study aims to develop a deep learning-based framework for automated learning behavior analysis to enhance classroom teaching evaluation. Our approach integrates Convolutional Neural Networks (CNNs) with multistage detection: Faster Region Based Convolutional Neural Networks (RCNN) localizes students, OpenPose extracts skeletal, facial and hand keypoints, and a CNN classifier categorizes behaviors. We define five key behaviors — *listening, fatigue, hand raising, sideways posture and reading/writing* — and introduce Upperclassmen, a curated dataset of 5,126 annotated images across these categories. Experimental results demonstrate robust performance, achieving 92.86% average validation accuracy, confirming suitability for real-time classroom deployment. Student learning behavior analysis in classrooms remains a challenge since human manual observation is time-consuming and subjective in nature. The study attempts to address the problem by proposing a deep learning-based framework consisting of Faster RCNN, OpenPose and the CNN10 classifier for real-time behavior recognition. By experimenting with the model, it demonstrates robust performance with an average accuracy of 97.92%, implying that this method is practical for real-life classroom monitoring.