I compared three statistical models — linear regression, random forest, and gradient boosting –with each achieving comparable accuracy. I ultimately selected the gradient boosting model, which slightly outperformed the other models, and included player height and weight as predictors (R² = 79%).
To rank 40-times, z-scores for each athlete were created by subtracting predicted 40-times from observed 40-times, with standard deviations tailored to the player’s weight range (for example, there was greater standard deviation in 40-times for 300 lb. players than 170 lb. players).
Feel free to email email@example.com for additional details.