Construction of a Data-Driven Prediction Model for Long Jump Performance and Design of Personalized Training Programs

This study constructs a data-driven prediction model for long jump performance by integrating kinematic analysis and machine learning. A dynamic data extraction model is first used to identify take-off and landing moments from videobased pose-estimation data, and the airborne trajectory is analyzed using a parabolic kinematic model. Multiple linear regression and random forest regression are then applied to quantify the influence of physical characteristics, including height, weight, and body fat percentage, together with technical variables such as take-off speed and take-off angle. The results identify take-off speed and angle as the most critical predictors of long jump performance. A personalized prediction case is further developed for a specific athlete, and targeted training interventions are proposed, including plyometric exercises and feedback-based technical drills to optimize the take-off angle. The study provides a systematic solution for performance evaluation, training optimization, and video-signal analysis in sports engineering, with potential extension to wireless sensing and real-time biomechanical monitoring.