Analyzing Athletes' Physical Performance and Trends in Athletics Competitions Using Time Series Data Mining Algorithms

Abstract

This study investigates the application of time series data mining algorithms to analyze athletes' physical performance and trends in athletics competitions. A cohort of 50 marathon runners was monitored over six months using wearable devices, including GPS trackers, accelerometers, and heart rate monitors. The collected data encompassed metrics such as speed, distance covered, heart rate, acceleration, stride length, and stride frequency, sampled at a frequency of 1 Hz. Preprocessing techniques, including noise removal, missing value imputation, and normalization, were applied to ensure data quality and consistency. Feature extraction methods were then employed to derive key performance attributes, such as mean speed, acceleration variance, and heart rate variability. Model training was conducted using Long Short-Term Memory (LSTM) networks, configured with two layers of 50 units each, to predict performance metrics and detect trends. The model's performance was evaluated using statistical metrics, including Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and R-squared (R²) score. Results indicate high predictive accuracy across multiple performance metrics, with R² scores exceeding 0.90 for speed and stride frequency predictions. Practical implications include the optimization of training regimens, race strategies, and injury prevention measures based on accurate performance predictions. Future research directions include expanding the scope to include diverse athlete populations and exploring other advanced machine learning algorithms to further enhance predictive accuracy and practical utility. Overall, this study highlights the potential of time series data mining algorithms in sports science for enhancing athletic performance monitoring and improvement.