Predicting Health-Related Physical Fitness From Smartwatch Data

Measuring health-related physical fitness (HRPF) is important due to its association with overall health, enabling disease prevention, personalized exercise prescription, among other applications. HRPF measurement is often performed on a laboratory setting, using clinical equipment under professional supervision, but has become more accessible due to wearable technology. For instance, current generation smartwatches are capable of predicting two of the five HRPF components acknowledged by the American College of Sports Medicine (ACSM), namely cardiorespiratory fitness (VO2max) and body composition (body fat %). In this paper, we present approaches for prediction of the remaining three HRPF components: muscular endurance, muscular strength, and flexibility, allowing a complete characterization of healthrelated physical fitness solely from smartwatch data. We adopted a component-specific approach, as opposed to more general approaches used in previous work, and show that our approach is more effective on data from wearable devices. Furthermore, we show that free-living activity data, exclusively accessible through wearables, is highly discriminative for HRPF prediction. This demonstrates that current consumer-grade wearables enable more than merely approximating clinical data — they also complement it with real-world activity information.