During college physical training, a substantial amount of data is continually generated through various activities. Historically, technological constraints have hindered the effective collection and utilization of this data, limiting its potential to enhance physical education (PE) and training through intelligent support. This study explores the application of data mining (DM) in managing and analyzing PE classrooms. It aims to streamline PE teaching management and evaluation, conduct in-depth analyses of students’ physical fitness data, and ultimately elevate the quality of school PE instruction. In designing a big data system for this purpose, data collection is prioritized based on its difficulty. To facilitate data analysis and presentation, a high-performance framework for storing and analyzing data is selected. The results indicate a 13.28% improvement in accuracy compared to traditional assessment models. By leveraging DM in PE classroom assessments, we can mitigate subjectivity and uncertainty associated with manual weight and correlation coefficient selection. This approach enhances the intelligence, adaptability, and usability of the assessment model, paving the way for more effective PE teaching and learning.