Scholarlink

<p>Physcial exercise and training are an integral part of athletes’ life. It defines the performance of the athletes based on their fitness and also helps in their mental wellbeing as well. The ability to monitor and predict training load helps to determine performance sustainability. Accurate analysis requires the integration of various factor such as biometric signals, physical exertion duration, recovery, and lifestyle features. This study proposes an attention-based TabTransformer model to predict training load by capturing contextual interactions between features. Top influential features have been identified using stadnard statistical methods like Principle Component Analysis, Uniform Manifold Approximation and Projection, Information Gain, and LASSO Regularization. The results reveals that the proposed TabTransformer outperofrms SOTA machine learning and deep learning baselines such as Random Forest, XGBoost, and Autoencoders. The empirical findings demonstrate that the TabTransformer explains the highest variance and provides the lowest error rates.</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

Atheletes’ training load prediction using transformers through data analytics and feature projection

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月22日

查看全文
  • Chen Tufa

摘要

Physcial exercise and training are an integral part of athletes’ life. It defines the performance of the athletes based on their fitness and also helps in their mental wellbeing as well. The ability to monitor and predict training load helps to determine performance sustainability. Accurate analysis requires the integration of various factor such as biometric signals, physical exertion duration, recovery, and lifestyle features. This study proposes an attention-based TabTransformer model to predict training load by capturing contextual interactions between features. Top influential features have been identified using stadnard statistical methods like Principle Component Analysis, Uniform Manifold Approximation and Projection, Information Gain, and LASSO Regularization. The results reveals that the proposed TabTransformer outperofrms SOTA machine learning and deep learning baselines such as Random Forest, XGBoost, and Autoencoders. The empirical findings demonstrate that the TabTransformer explains the highest variance and provides the lowest error rates.