Toward detecting the zone of elite tennis players through wearable technology

Abstract

Wearable devices fall short in providing information other than physiological metrics despite athletes' demand for psychological feedback. To address this, we introduce a preliminary exploration to track psychological states of athletes based on commercial wearable devices, coach observations and machine learning. Our system collects Inertial Measuring Unit data from tennis players, while their coaches provide labels on their psychological states. A recurrent neural network is then trained to predict coach labels from sensor data. We test our approach by predicting being in the zone, a psychological state of optimal performance. We conduct two experimental games with two elite coaches and four professional players for evaluation. Our learned models achieve above 85% test accuracy, implying that our approach could be utilized to predict the zone at relatively low cost. Based on these findings, we discuss design implications and feasibility of this approach by contextualizing it in a real-life scenario.