Maximum heart rate (MHR) is widely used in the prescription and monitoring of exercise intensity,
and also as a criterion for the termination of sub-maximal aerobic fitness tests in clinical populations.
Traditionally, MHR is predicted from an age-based formula, usually 220-age. These formulae, however, are
prone to high predictive errors that potentially could lead to inaccurately prescribed or quantified training
or inappropriate fitness test termination. In this paper, we used functional data analysis (FDA) to create
a new method to predict MHR. It uses heart rate data gathered every 5 seconds during a low intensity,
sub-maximal exercise test. FDA allows the use of all the information recorded by monitoring devices in the
form of a function, reducing the amount of information needed to generalize a model, besides minimizing the
curse of dimensionality. The functional data model created reduced the predictive error by more than 50%
compared to current models within the literature. This new approach has important benefits to clinicians and
practitioners when using MHR to test fitness or prescribe exercise.
Keywords: Maximum heart rate prediction, functional data analysis, machine learning, low intensity sub-maximal test