<p>The impact of students’ physical behaviors in the classroom on learning gains is well-documented; however, manual analysis of such behaviors from classroom videos is labor-intensive and time-consuming. Recent advancements in deep learning technologies have enabled the automatic detection and analysis of students’ physical behaviors and their correlation with learning gains. Despite these advancements, the exploration of the intricate relationship between physical behaviors and learning gains remains underdeveloped. This study introduces a novel analytical approach that utilizes deep learning techniques to investigate this relationship. Initially, a deep learning-based algorithm was employed to detect and classify students’ physical behaviors from video recordings in smart classrooms. Learning gains were assessed using pre-test and post-test data. A dataset was constructed for correlation analysis, with experiments conducted from two perspectives: tracking individual students across multiple classes and examining the relationship between classroom behaviors and learning gains for all students in a particular class. Bivariate correlation analysis was utilized to evaluate the association between physical behaviors and learning gains. The findings demonstrate that deep learning methods are effective for automating the recognition of students’ physical behaviors and reveal a significant correlation between these behaviors and learning gains. This study provides important insights into the use of artificial intelligence and statistical methods for educational learning analysis, thereby opening new avenues for research in this area.</p>