Predicting 2000-Meter Indoor Rowing Performance Using Accessible Machine Learning Models

Abstract

The 2000-meter ergometer test is widely used to measure athlete's strength, skill and efficiency in competitive rowing. Traditional tests like 500m or 1000m rowing can be too physical and elaborate for beginners or younger rowers. This study aimed to create a simpler and data driven approach to predict 2000m rowing times using basic information like age, gender and weights. Predictions were made using machine learning models including XGBoost that were applied to data from 1,341 rowers obtained from Concept 2 Database and Miami Beach Rowing Club. The model performed better for athletes over 18 years old with gender as the most important factor followed by weight and age. Finally after rigorous model training, the model showed insightful prediction accuracy with R2=0.75, MAE=0.35 min and RMSE=0.47 min. However, cross-validation of the model showed R2=-2.04, indicating overfitting due to limited variables and data. Despite this limitation, our model offers a practical application that can help rowers set realistic goals and assist coaches in personalized training. In conclusion, the model can still be improved to improve accuracy and validation but in the current study it represents a step forward in making performance insights more accessible to rowers.