06/21/2024
Another study from my PhD just came out!
Predicting daily recovery during long‐term endurance training using machine learning analysis - a study that looked to predict daily recovery using training metrics, diet, sleep, HRV, and subjective well-being.
Coaches and athletes commonly measure so many things, what actually matters most to inform recovery? I had 55 athletes record everything for 12 weeks (3572 days of tracking). In short, a small subset of variables provide most of the predictive power at the group level.
At the individual level, predictive accuracy varies greatly as do the variables with the most predictive power. However once you find them for each athlete, a small subset of variables offers similar predictive capability as the full dataset.
Of interest to the machine learning folks, the LASSO model outperformed all of the fancier ones at the group level. For individuals the winners were LASSO: 30%, SVM: 30%, XGBoost: 23%, Light GBM: 12%, KNN: 4%, Ridge: 1%, and MARS: 0.4% of models.
Future research in single-sport athletes would allow additional load metrics like HR, total work, or distance to be more easily factored in. Also energy availability, alcohol intake, and menstrual cycle tracking could be helpful metrics to include.
Free to download here -
https://link.springer.com/article/10.1007/s00421-024-05530-2
