Machine Learning Estimation of COVID-19 Social Distance Using Smartphone Sensor Data

Semenov, Oleksandr

Etd

Machine Learning Estimation of COVID-19 Social Distance Using Smartphone Sensor Data

Public

COVID-19 is spread from an infected to a healthy person when they are within 6 feet from each other for longer than 15 minutes. To limit disease transmission, there is a need for technology that could identify whether subjects were near each other longer than 15 minutes. In this thesis we systematically investigate Machine Learning (ML) methods to detect proximity by analyzing data gathered from smartphones’ built-in Bluetooth, accelerometer, and gyroscope sensors. We show that the proximity classification (< 6ft or not) can achieve 72%-90% accuracy using the accelerometer, 78%-84% accuracy using gyroscope sensor, and 76%-92% accuracy with the Bluetooth radio, while sensor fusion shows accuracy as high as 97%. Our model outperforms current state-of-the-art methods using neural networks and achieved Normalized Decision Cost Function (nDCF) score of 0.34 with Bluetooth radio and 0.36 with sensor fusion.

Creator