Improving Tour de France modeling with allometric scaling

Abstract

We describe improvements we have made to the model our research group employs to predict stage-winning times of the Tour de France. Accounting for different stage-winning cyclist masses associated with different stage types, we use allometric scaling to modify our model’s cyclist power output. We show definitively that such a change to our model improved our prediction capabilities over what we were able to predict using the model we employed for the 2013 Tour de France. Excluding three tailwind-dominated stages, where our worst error was 7.79%, we predicted all other stages to better than 5%, including five stages that we predicted to better than 1%. We also show how to improve our model further with a different type of scaling.