Prediction of Marathon Performance using Artificial Intelligence.

Area of Science:

• Sports Science
• Data Science
• Biomechanical Engineering

Background:

• Machine learning is increasingly applied to predict athletic performance.
• Few studies have validated artificial intelligence (AI) techniques like artificial neural networks (ANN) and k-nearest neighbors (KNN) specifically for marathon running.
• There's a need to compare the predictive accuracy of different AI models in endurance sports.

Purpose of the Study:

• To validate and compare the performance prediction accuracy of ANN and KNN algorithms in marathon running.
• To assess the reliability of AI-driven marathon time estimations using athlete data.
• To determine which AI technique, ANN or KNN, offers superior precision for predicting marathon race outcomes.

Main Methods:

• Utilized official French 10-km and marathon rankings from 2019, comprising 820 athletes.
• Employed ANN and KNN models with identical input variables: 10-km race time, BMI, age, and sex.
• Applied a linear regression approach to estimate marathon race times based on the selected input parameters.

Main Results:

• Both ANN and KNN models accurately predicted marathon performances, showing no significant difference from actual race times (p>0.05).
• All predicted performances exhibited strong correlations with actual results (r>0.90; p<0.001).
• K-nearest neighbor (KNN) outperformed artificial neural network (ANN), achieving a lower mean absolute error (2.4% vs. 5.6%).

Conclusions:

• Both artificial intelligence algorithms (ANN and KNN) are valid tools for predicting marathon performance.
• K-nearest neighbor (KNN) provides higher accuracy compared to artificial neural network (ANN) in marathon time prediction.
• AI-based performance predictions can be effectively integrated into athletic training regimens and competition strategies.