A Game-Theoretical Approach for Data Acquisition From Fitness Tracking Devices

Fitness and wellness apps and devices for tracking physical activities have proliferated in the recent years, enabling access to large data sets collected in the free-living environment. Such data sets may provide insights into human physical and health behaviors, and they may be used to model a specific physical response or performance outcome. Increasingly, statistical machine learning is used to derive predictors of human physical performance from these large data sets, replacing traditional physics-based approaches. However, not all acquired data are useful for machine learning, and quantification of the value of the data is critical for data acquisition in the Internet of Things fitness applications. This paper presents a game-theoretical approach for the valuation of data collected by cycling devices and apps. We demonstrate the utility of this framework using experiments with real-world cycling data sets and show that by acquiring high-valued data, existing predictors of cycling performance can be improved. In contrast, low-valued data decrease their accuracy. The proposed approach is simple to implement, and it can be readily adopted for the acquisition of new data from any fitness app or device.