Sports data analysis of track and field athletes based on IID-OPK and NonIID-OPK algorithms

The traditional K-means clustering algorithm randomly selects clustering center points, which can easily lead to unstable clustering results and be affected by outliers, making it unable to handle complex sports data of track and field athletes. Therefore, the research introduces the Independent and Identically Distributed (IID) and Non-Independent and Identically Distributed (NonIID) ideas based on the K-means algorithm to optimize the data analysis process. The study used a training dataset from a track and field team under the Shandong Provincial Sports Bureau in China over the past two years (2022–2024). This includes basic physical fitness data (such as height 165–195 cm, weight 48–85 kg, lung capacity 3500–6500 ml), speed and strength training data (such as 9.58–12.50 s for 100 m sprint 13:50–18:30 for 5000 m long-distance running), as well as technical and exercise data collected through inertial measurement units. The experimental results showed that the K-means model (NonIID-OPK) optimized based on the NonIID assumption had the best performance, with an accuracy of 96.9 %, SSE value of 68.1, F1 value of 97.2 %, recall rate of 96.8 %, execution time of 44.3 s, iteration times of 0.84, accuracy of 92.7 %, and contour coefficient of 0.64. Compared with the K-means Optimization based on Independent and Identically Distribution data (IID-OPK), the NonIID-OPK model performed better in clustering performance and computational efficiency. In the case analysis, the optimal K value of the model was 6, and the branch contour coefficient after clustering was the highest, indicating that it was more discriminative in classifying athlete data. Further application shows that after guiding training based on the analysis results, athletes' long-distance running performance, short distance running performance, lung capacity, and other indicators have significantly improved. Overall, the NonIID-OPK algorithm can effectively explore potential patterns in athletic data of track and field athletes, providing reliable support for scientific training.