Machine Learning Framework for Standing Long Jump Assessment and Training Optimization

Abstract

With the implementation of the "National student physical fitness standards," standing long jump physical assessment faces challenges such as large manual assessment errors, inaccurate movement diagnosis, and a lack of personalized training recommendations. This paper aims to utilize AI technology to construct a multidimensional modeling system to accurately analyze and predict standing long jump performance. Specifically, using frame-level coordinate data from 33 key human nodes, the study firstly identifies takeoff and landing moments through data preprocessing and a linear Support Vector Machine (SVM) model, quantifying movement characteristics such as arm swing amplitude and joint angles during the hovering phase. Secondly, combining grey correlation analysis with an XGBoost regression model, six key influencing factors, including muscle rate, hovering time, and takeoff angle, were identified. A gradient boosting regression tree model was constructed to predict the performance of candidate athletes. Finally, a phased short-term training plan was designed based on a constrained optimization model, estimating that the ideal performance could be improved significantly. This study provides a machine learning framework for AI-assisted physical assessment, from movement recognition to personalized training, and promoting the intelligent transformation of youth physical fitness monitoring.