Scholarlink

<p>The aim of this study was to evaluate whether machine learning models comprising anthropometric, physical fitness, and sport-specific performance data from young canoe sprint athletes can predict their senior performance level (SPL). Between 1992 and 2019, anthropometric (e.g. body mass/height), physical fitness (e.g. 800 m/1500 m run, 2&#xa0;min bench press/pull), and sport-specific performance (e.g. 250 m/2000&#xa0;m on-water canoe sprint) data as well as age (i.e. U13 to U16) and sport discipline were annually examined in young male and female canoe sprint athletes (n = 729, male: 495, female: 234). A benchmark experiment was conducted to evaluate and compare multiple classification models and to use the final model to predict SPL (national vs. international) on three validation datasets (n = 103, U13 to U17, 2021 to 2023) with ground-truth labels from 2025. Findings revealed that an XGBoost model achieved acceptable discrimination (AUC = 0.81) and balanced accuracy (BACC = 0.73) for predicting SPL in young canoe sprint athletes. However, precision for the international class was low (PRAUC = 0.35, PPV = 0.20), indicating many false-positive international predictions. The most important feature was 2000&#xa0;m on-water canoe sprint test. Furthermore, predictions on the three external validation datasets showed limited temporal generalizability, with moderate discrimination (AUC: 0.68 to 0.73), modest but consistently above-chance balanced accuracy (BACC: 0.59 to 0.63), and moderate but variable sensitivity (0.20 to 0.67). However, precision for identifying international athletes was low across the external validation datasets, indicating a high false-positive rate. Therefore, the model should be interpreted as an acceptable screening tool. However, low precision and variable sensitivity limit its practical utility as a stand alone selection instrument. The present findings may help practitioners involved in talent selection and development in Olympic canoe sprinting and may inform the development of future prediction models for young canoeists based on anthropometric, physical fitness, and sport-specific performance data.</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

Anthropometrics, physical fitness, and sport-specific performance of young German canoe sprint athletes (U13-U17) to predict senior performance level: a machine-learning approach

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月21日

查看全文
  • Christian Saal,
  • Jan Willem Teunissen,
  • Urs Granacher,
  • Norman Helm,
  • Torsten Warnke,
  • Olaf Prieske

摘要

The aim of this study was to evaluate whether machine learning models comprising anthropometric, physical fitness, and sport-specific performance data from young canoe sprint athletes can predict their senior performance level (SPL). Between 1992 and 2019, anthropometric (e.g. body mass/height), physical fitness (e.g. 800 m/1500 m run, 2 min bench press/pull), and sport-specific performance (e.g. 250 m/2000 m on-water canoe sprint) data as well as age (i.e. U13 to U16) and sport discipline were annually examined in young male and female canoe sprint athletes (n = 729, male: 495, female: 234). A benchmark experiment was conducted to evaluate and compare multiple classification models and to use the final model to predict SPL (national vs. international) on three validation datasets (n = 103, U13 to U17, 2021 to 2023) with ground-truth labels from 2025. Findings revealed that an XGBoost model achieved acceptable discrimination (AUC = 0.81) and balanced accuracy (BACC = 0.73) for predicting SPL in young canoe sprint athletes. However, precision for the international class was low (PRAUC = 0.35, PPV = 0.20), indicating many false-positive international predictions. The most important feature was 2000 m on-water canoe sprint test. Furthermore, predictions on the three external validation datasets showed limited temporal generalizability, with moderate discrimination (AUC: 0.68 to 0.73), modest but consistently above-chance balanced accuracy (BACC: 0.59 to 0.63), and moderate but variable sensitivity (0.20 to 0.67). However, precision for identifying international athletes was low across the external validation datasets, indicating a high false-positive rate. Therefore, the model should be interpreted as an acceptable screening tool. However, low precision and variable sensitivity limit its practical utility as a stand alone selection instrument. The present findings may help practitioners involved in talent selection and development in Olympic canoe sprinting and may inform the development of future prediction models for young canoeists based on anthropometric, physical fitness, and sport-specific performance data.