How to Deal with Inaccurate Sensor Data?

A lot of domains and applications rely on high quality sensor data. In chemical processes, the stability of the process is critical, while in mobile health apps the well-being of the patient is being involved. Contrary to chemical processes, mobile health apps feature a lot of sensors as it is very easy to collect huge amounts of sensor data using techniques like mobile crowdsensing. For example, the Apple Heart Study has collected the data of over 400,000 users. In mobile health apps, sensor faults are more likely due to the larger number of sensors and they are harder to predict because each user has their own device with a potentially different kind of sensor like it is the case with TYT. In this work, we are going to primarily focus on the domain of mobile health apps that are powered by mobile crowdsensing like the Apple Heart Study or TYT. Studies and research on the sensed data allow the processing of large amounts of data that would otherwise be very difficult with a conventional study. However, this flexibility also comes with a cost. While in a conventional study, every participant can use the same measurement device, a study based on mobile crowdsensing involves a huge number of different device types. As previously told, these different device types reduce the predictability of sensor faults. That’s why our goal should be to develop an algorithm, which can detect and identify sensor faults in mobile health data sets. Devices, where the faults are too severe, should be excluded from the data set. In addition to the sensed data, the algorithm should also use additional information that is gathered along with the sensor data. For example, this extra information are the answers to the questions that rate the tinnitus severity in the case of TYT.