Risk Estimation in COVID-19 Contact Tracing Apps

Abstract

In the context of COVID-19, contact tracing has shown its value as a tool for contention of the pandemic. In addition to its paper based form, contact tracing can be carried out in a more scalable and faster way by using digital apps. Mobile phones can record digital signals emitted by communication and sensing technologies, enabling the identification of risky contacts between users. Factors such as proximity, encounter duration, environment, ventilation, and the use (or not) of protective measures contribute to the probability of contagion. Estimation of these factors from the data collected by phones remains a challenge. In this work in progress we describe some of the challenges of digital contact tracing, the type of data that can be collected with mobile phones and focus particularly on the problem of proximity estimation using Bluetooth Low Energy (BLE) signals. Specifically, we use machine learning models fed with different combinations of statistical features derived from the BLE signal and study how improvements in accuracy can be obtained with respect to reference models currently in use.