This article was accepted into the “Learning from Time Series for Health” workshop at NeurIPS 2022.
Heart rate (HR) dynamics in response to training intensity and duration measures key aspects of a person’s cardiorespiratory health and fitness. Exercise physiology models have been used to characterize cardiorespiratory fitness in well-controlled laboratory settings, but face additional challenges when applied to wearable devices in noisy real-world environments. Here, we present a hybrid machine learning model that combines a physiological model of HR and exercise demand with neural network embeddings to learn user-specific fitness parameters. We apply this model to scale to a large set of training data collected with wearable devices. We show that this model can accurately predict the heart rate response to exercise demand in new training. Furthermore, we show that learned intakes correlate with traditional metrics that reflect cardiorespiratory fitness.
