Design and Implementation of Street-level Crowd Density Forecast using Contact Tracing Applications

Abstract

Social distancing plays an important role in the control of the spread of infectious diseases. This study proposes a service that forecasts street-level crowd density in the near future. We collected street-level crowd density levels for months during the COVID-19 pandemic by observing public Bluetooth Low Energy advertisements from popular contact tracing applications. We then designed a model to predict crowd density level from other factors such as calendars, weather, and recent trends of crowd density level using Random Forest Regressor. Based on the model, we implemented a crowd density forecast service by incorporating an external weather forecast service, and we published the forecast on our website and a Japanese television program. The experimental results indicate that the model can predict the crowd density for the following week with a coefficient of determination of 0.85 or higher on average, which demonstrates that a practical crowd density forecast can be realized with our method.