Exploration of machine learning based on big data in sports models and physical education teaching

Article ID: 940

Abstract

Under the background of big data-driven education digitization, physical education in colleges and universities is facing the problems of data isolation (< 35% compatibility rate of multi-source devices) and training homogenization (only a 28% coverage rate of personalized programs). In this study, we integrated six existing data specifications, including the Student Physical Fitness Standard, and designed a standardized data collection scheme for physical fitness monitoring: (1) We developed a data meta-standard that included 12 types of wearable devices (heart rate/step rate/oxygen uptake) and three types of training scenarios (classroom/outside classroom/competition) and unified the collection protocols for the 23 core metrics; (2) We adopted a machine-learning preprocessing method K-Nearest Neighbors (KNN) filling of missing values + Isolation Forest detection of outliers) to make the data standardized. (3) Using machine learning preprocessing (missing value KNN filling + anomaly detection by Isolation Forest), the data completeness rate is increased from 61% to 89%; (4) Constructing a comprehensive physical fitness scoring model (XGBoost algorithm, with 8 basic physical fitness and 5 dynamic response indicators) and combining it with the reinforcement learning recommender engine, generating a hierarchical teaching plan (with 3 levels of difficulty adaptation and 5 types of sports intervention templates) for teachers. Empirical evidence shows that the program improves the efficiency of teachers’ lesson planning by 46.42% and increases the matching degree of students’ personalized training by 3.2 times. Compared with traditional empirical teaching, the standardized data-based intelligent model reduces the error rate of physical fitness assessment from 19.7% to 8.3% and provides teachers with a three-in-one tool: class heat map—individual warning list—lesson plan knowledge base, which pushes the physical education teaching from “experience-driven” to “data-driven”. This study provides a reusable methodological framework for the digital transformation of college sports.