Scholarlink

<p>This paper presents an integrated software and hardware solution leveraging machine learning (ML) techniques and wearable sensor technologies to automate athletic talent assessment. Referred to as the Machine Learning-based Athlete Talent Recognition (ML-ATR) system, ML-ATR addresses limitations of traditional scouting, which relies on subjective evaluations and lacks real-time performance assessment, by introducing a gesture pattern refinement process to enhance motion data consistency and improve classification reliability. Applied to football shooting talent recognition, ML-ATR initially achieved 91% accuracy using Support Vector Machine (SVM) on sensor data. However, overfitting concerns necessitated sliding window data augmentation, which led to an accuracy improvement of 96%. The comparative analysis confirmed SVM’s superiority over K-Nearest Neighbors (KNN), Random Forest, Decision Tree, and Naive Bayes in generalization performance. Although ML-ATR was only tested on a football shooting case study, unlike conventional methods that rely on video analytics or subjective assessments, ML-ATR offers real-time evaluation and adaptability across multiple sports. Its structured refinement process enhances motion data reliability, making it applicable to many other sports such as basketball, tennis, and sprinting.</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

Adaptive Multi-Sport Smart Tracker for Athlete Talent Identification

  • 期刊论文

  • 发表日期：

    2025年4月26日

查看全文
  • Khaled Necibi,
  • Ahmed-Chawki Chaouche,
  • Zakaria Soukeur

摘要

This paper presents an integrated software and hardware solution leveraging machine learning (ML) techniques and wearable sensor technologies to automate athletic talent assessment. Referred to as the Machine Learning-based Athlete Talent Recognition (ML-ATR) system, ML-ATR addresses limitations of traditional scouting, which relies on subjective evaluations and lacks real-time performance assessment, by introducing a gesture pattern refinement process to enhance motion data consistency and improve classification reliability. Applied to football shooting talent recognition, ML-ATR initially achieved 91% accuracy using Support Vector Machine (SVM) on sensor data. However, overfitting concerns necessitated sliding window data augmentation, which led to an accuracy improvement of 96%. The comparative analysis confirmed SVM’s superiority over K-Nearest Neighbors (KNN), Random Forest, Decision Tree, and Naive Bayes in generalization performance. Although ML-ATR was only tested on a football shooting case study, unlike conventional methods that rely on video analytics or subjective assessments, ML-ATR offers real-time evaluation and adaptability across multiple sports. Its structured refinement process enhances motion data reliability, making it applicable to many other sports such as basketball, tennis, and sprinting.