The aim of this research was to develop a physical education teaching model based on machine learning. This research is based on the philosophy (paradigm) of pragmatism, qualitative in terms of method and data-driven in terms of strategy. Participants included experts in the research topic, including faculty members of physical education and sports sciences faculties across the country who have at least the rank of associate professor or professor and have more than 10 years of teaching experience. The sampling method in the qualitative part of the research was conducted using two purposeful and theoretical saturation interviews with 13 people. According to Guba and Lincoln (2000), the validity of the research included four criteria of validity, transferability, reliability and confirmability. The Lavish method was used to assess the reliability of the interview protocol. In this research, the opinions of 11 experts were obtained for validation using the Lavish method, which was finally calculated based on the results of 5 item scores less than 0.6, so these items were removed from the model and a 65-component model was extracted. Based on the findings, the results of a 13-factor model with 65 factors in the form of a data-driven model with the Strauss and Corbin approach were presented for the implementation of physical education teaching based on machine learning. Using machine learning algorithms, it is possible to predict contextual patterns and algorithms, intervening factors, strategies, and outcomes as elements of the paradigm model. In summary, the use of machine learning in physical education teaching not only increases the quality of education, but also facilitates the process of evaluation and feedback, optimization of educational resources, and personalization of learning for students. These advances can play an important role in improving students' sports skills and promoting their physical and mental health.