This is my PhD thesis. It wraps all related publications into one narrative.
Summary: Data science advances in sports commonly involve ‘big data’, i.e., large sport-related data sets. However, such big data sets are not always available, necessitating specialized models that apply to relatively few observations. One important area of sport science research that features small data sets is the study of energy recovery during intermittent exercise. In this area, models are typically fitted to data collected from exhaustive exercise test protocols, which athletes can perform only a few times.
Recent findings highlight that established recovery models, such as the so-called work-balance models, are too simple to adequately fit observed trends in the data. These models summarize the available energy capacities of an athlete during exercise in a single variable, which is referred to as work balance. In this thesis we revisit a so-called hydraulic performance model and hypothesize that it is able to address the recently highlighted shortcomings of work-balance models. However, current literature has not fully validated the original hydraulic model, because it depends on physiological measures that cannot be acquired at the required precision or quantity.
We introduce a generalized interpretation and formalization of the original hydraulic model that removes its ties to concrete physiological measures. We use evolutionary computation to fit its parameters to an athlete. In this way, we investigate a new hydraulic model that requires the same few data points as work-balance models, but promises to predict recovery dynamics more accurately.
To compare the hydraulic model to established work-balance models, we retrospectively apply them to data compiled from previously published studies. The hydraulic model outperforms established work-balance models on all defined metrics, even those that penalize models featuring higher numbers of parameters. However, the more accurate energy recovery predictions of the hydraulic model come at the cost of inaccurate predictions of metabolic responses during exercise, such as oxygen uptake.
In conclusion: While the new hydraulic performance model should not be used to predict metabolic responses during exercise, it promises to be a powerful tool for predicting energy recovery. This work carefully positions the new hydraulic model among existing models, with its benefits and limitations. The results render the new hydraulic model a powerful alternative to address the shortcomings of established work-balance models and incentivize further investigation. Data and code are published as open source.