Precision of human workout-time detection using Random Forests and Wearable Sensor Data

The widespread use of wearable sensor technology has made it possible to obtain a variety of human biological information. Among them, workout is important for health promotion, and estimation of exercise duration and intensity is a clear and convenient way to understand health status. Therefore, it is desirable to be able to estimate workout efficiently. Many existing wearable sensors can measure the accumulated intensity of aerobic exercise using heart rate or provide a rough estimate. However, the estimation algorithm has not been published, and it is not clear how accurate the workout can actually be detected. In this study, we attempted to detect workout time from biometric data obtained over a long period of time using random forest. The results showed a high estimation, with 0.96 accuracy and 0.92 recall. As a result, workout was considered easy to estimate.