Construction of a neural network-based model for training data analysis and performance prediction of athletes

• Published in: Applied mathematics and nonlinear sciences Vol. 10; no. 1
• Main Author: Xie, Shiyu
• Format: Journal Article
• Language: English
• Published: Beirut Sciendo 01.01.2025; De Gruyter Poland
• Subjects: 68T01; Athlete training data analysis; Athletes; Automatic codec network model; Clustering; Dynamic time regularization algorithm; KinectV2; Self-organizing mapping network
• ISSN: 2444-8656; 2444-8656
• DOI: 10.2478/amns-2025-0624

At present, the traditional sports training management mode is obsolete, which can’t do the scientific guidance for athletes’ training. This paper proposes a movement prediction system based on self-organizing mapping network, which firstly adopts a filter combining jitter clear and bi-exponential smoothing to filter the skeletal data of athletes collected by KinectV2. Then an automatic coding and decoding network model is designed, which accomplishes the task of extracting athlete-related motion information by dividing human limb parts and extracting independent features of each part. The extracted athlete movement information is matched with the standard movement template by combining the dynamic time regularization algorithm with the Euclidean distance to achieve the evaluation of the athlete’s training data, and finally, the self-organizing mapping network is introduced to predict the clustering of the athlete’s training performance. The score reliability of this paper’s algorithm is above 97%, and the self-reported movement states of the athletes participating in the test are consistent with the training state clustering results under the cascade self-organizing mapping network clustering. It shows that the model established in this paper is characterized by high accuracy, objectivity, and scientificity, which can accurately predict the performance of athletes and provide scientific training guidance.