Predicting Elite 100m Sprint Performance: A Comparative Analysis of Regression and Machine Learning Model

Abstract

Sprint performance can be analyzed through a data-driven approach by utilizing biomechanical parameters that play an important role in athletic development. The two biomechanical parameters explored in this study are Cadence and step length which are analyzed for their influence on running speed. Data was taken from six elite world sprinters in the men's 100 meter final of the 2020 Tokyo Olympics focusing on performance dynamics at every 10 meter interval to capture the real performance of athletes in competition. The modeling approach uses regression and machine learning, namely Random Forest (RF) and Neural Network (NN). Linear regression is used as a basic model that is easy to interpret, while Random Forest and Neural Network are used in the digital-based automated performance analysis system. These three models will be evaluated using the coefficient of determination (R²) and error (MSE) to compare their accuracy. The results show that all three models are capable of providing high accuracy in predicting running speed. The regression model provides strong statistical validity with R² of 0.9217 and MSE of 0.0062. The Random Forest model produced the highest prediction accuracy with R² of 0.9757 and MSE of 0.1027. Meanwhile, the Neural Network offers a more flexible approach in capturing complex patterns from the data with R² of 0.9261 and MSE of 0.3117. These results confirm the importance of integrating digital technology and intelligent predictive models in optimizing athletic performance. Thus, training programs can be made more specific and data-driven, improving the efficiency and effectiveness of athlete performance development.