Scholarlink

<p>This study evaluated a GAN-inspired predictive-generative decision-support framework for individualized training-load prescription using longitudinal physiological monitoring data from elite athletes. The framework combined three components: a discriminator for adaptive-state estimation, a generator for candidate seven-day training prescriptions, and a forward dynamics model for simulated physiological response forecasting. A total of 128 elite athletes from multiple sport categories were followed over 16 months, including a 10-month model-development phase and a 6-month coach-supervised validation phase. The validation design was non-randomized; therefore, between-group findings were interpreted as adjusted associations rather than causal treatment effects. Multimodal inputs included training-load measures, heart rate variability, sleep indicators, neuromuscular performance, subjective wellness, biochemical markers, and sport-specific performance outcomes. Internal computational validation showed acceptable overall forward-dynamics prediction performance, although biochemical markers remained the least predictable physiological domain. Expert evaluation indicated that generated prescriptions were rated within a comparable practical range to human-designed prescriptions, but source-identification confidence was moderate, supporting cautious interpretation. During internal validation, the model-supported workflow was associated with more favorable training efficiency, lower overtraining-marker burden, improved standardized performance trajectories, and lower non-contact injury incidence compared with conventional coach-directed periodization. However, contact injury rates showed no meaningful between-group difference, and baseline comparisons remained limited to the internal dataset and evaluation protocol. These findings suggest that GAN-inspired decision support may be useful for structuring individualized training-load decisions under coach supervision. Nevertheless, the single-cohort design, lack of randomization, absence of external validation, interpretability constraints, and deployment requirements limit generalizability. Randomized controlled evaluation and independent multi-site validation are essential before broader applied implementation can be supported.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A generative adversarial network framework for individualized training load distribution based on physiological response patterns and injury risk prediction

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月28日

查看全文
  • Qingbin Chen,
  • Samir Karaman

摘要

This study evaluated a GAN-inspired predictive-generative decision-support framework for individualized training-load prescription using longitudinal physiological monitoring data from elite athletes. The framework combined three components: a discriminator for adaptive-state estimation, a generator for candidate seven-day training prescriptions, and a forward dynamics model for simulated physiological response forecasting. A total of 128 elite athletes from multiple sport categories were followed over 16 months, including a 10-month model-development phase and a 6-month coach-supervised validation phase. The validation design was non-randomized; therefore, between-group findings were interpreted as adjusted associations rather than causal treatment effects. Multimodal inputs included training-load measures, heart rate variability, sleep indicators, neuromuscular performance, subjective wellness, biochemical markers, and sport-specific performance outcomes. Internal computational validation showed acceptable overall forward-dynamics prediction performance, although biochemical markers remained the least predictable physiological domain. Expert evaluation indicated that generated prescriptions were rated within a comparable practical range to human-designed prescriptions, but source-identification confidence was moderate, supporting cautious interpretation. During internal validation, the model-supported workflow was associated with more favorable training efficiency, lower overtraining-marker burden, improved standardized performance trajectories, and lower non-contact injury incidence compared with conventional coach-directed periodization. However, contact injury rates showed no meaningful between-group difference, and baseline comparisons remained limited to the internal dataset and evaluation protocol. These findings suggest that GAN-inspired decision support may be useful for structuring individualized training-load decisions under coach supervision. Nevertheless, the single-cohort design, lack of randomization, absence of external validation, interpretability constraints, and deployment requirements limit generalizability. Randomized controlled evaluation and independent multi-site validation are essential before broader applied implementation can be supported.