Prediction Model for Standing Long Jump Performance Constructed Using Lasso Regression

Abstract

Against the backdrop of promoting the National Student Physical Fitness Standards, this paper proposes an AI posture analysis and evaluation framework for the standing long jump. It identifies key events from keypoint sequences and explains performance variations. The movement process is divided into three phases: takeoff, flight, and landing. Posture features are constructed based on trajectories of 33 body keypoints. Collected data undergoes cleaning and smoothing, removing 438 anomalous samples with all zeros and performing mean imputation on 26 samples. Kinematic representations are extracted via least-squares quadratic trajectory fitting during the flight phase. Lasso regression establishes relationships between posture variables and jump distance to screen critical motion features. Taking Athlete 1 as an example, the model predicted a performance of 1.545m and provided a short-term training target of 1.676±0.068m. This framework can be applied to physical assessment and training focus determination, with further validation required on larger samples and diverse body types.