AI-Driven Personalized Training Recommender for Adolescent Sports: A Multimodal Adaptive Learning Framework

Abstract

This paper presents an AI-driven personalized training recommendation system designed for the unique needs of youth sports training. It addresses the variability of physiological and biomechanical attributes of young athletes. To provide adaptive training recommendations, the framework integrates a bi-directional long short-term memory network (Bi-LSTM) for modeling the temporal dynamics of exercise sequences; a convolutional neural network (CNN) for extracting spatial features related to skeletal biomechanics; and hierarchical reinforcement learning. The dynamic reward function balances athletic performance improvement and safety by using sport-specific metrics such as skill improvement rate and injury risk. A Bayesian neural network enhances the system through permissible reminder parameter updates. The system was tested on a dataset of 1852 youth athletes participating in basketball and track and field programs. Compared to the baseline model, the system improved recommendation accuracy by 18.9 % and reduced the risk of overtraining-related injuries by 31.2 %. The system demonstrated its practical applicability in dynamic sports environments, with real-time feedback and an inference latency of less than 200 milliseconds. The framework leverages multimodal data, including physiological signals and movement trajectories, to provide a scalable, adaptive, and efficient solution to maximize performance and reduce injury risk. This research provides a solid foundation for an intelligent athletic training system that significantly improves the development of youth sports in terms of safety, precision and efficiency.

Ähnliche Arbeiten