Forecasting Exercise Exertion Levels Using LSTM Modeling of Wearable Physiological Data

This pilot study aimed to predict exercise exertion levels from physiological data collected from wearable devices using Long Short-Term Memory (LSTM) Recurrent Neural Network (RNN). During a 16-minute cycling exercise, real-time ECG, pulse rate, oxygen saturation, and revolutions per minute (RPM) data were collected at three intensity levels. In addition, the study subjects' ratings of perceived exertion (RPE) were gathered once per minute throughout each exercise session. Each 16-minute exercise session was further divided into eight 2-minute windows. Based on the self-reported RPEs, each 2-minute window was labeled as either “high exertion” or “low exertion” classes. RPMs, heart rate, and oxygen saturation levels were averaged for each window to form the predictive features. In addition, collected ECG data were used to extract the heart rate variability (HRV) features in both the temporal and frequency domains. We used the minimum redundancy maximum relevance (mRMR) algorithm to select the best predictive features. The top selected features were then used to train and test the LSTM classifier to predict the next window's exertion level. The designed classifier showed the training accuracy of 85.7% and the testing accuracy of 78.6%. This pilot study shows the potential of a deep learning classifier to track perceived exercise exertion automatically in real-time.