Estimating Attendance From Cellular Network Data

TL;DR

This paper introduces a method to estimate event attendance in cities using anonymized cellular network data, achieving median errors below 15% across various event types.

Contribution

The authors develop a novel approach combining network cell identification and user activity verification to accurately estimate event attendance from CDR data.

Findings

Median error in attendance estimation is less than 15%.

Method works across different event types like sports, concerts, and festivals.

Approach effectively distinguishes attendees from non-attendees.

Abstract

We present a methodology to estimate the number of attendees to events happening in the city from cellular network data. In this work we used anonymized Call Detail Records (CDRs) comprising data on where and when users access the cellular network. Our approach is based on two key ideas: (1) we identify the network cells associated to the event location. (2) We verify the attendance of each user, as a measure of whether (s)he generates CDRs during the event, but not during other times. We evaluate our approach to estimate the number of attendees to a number of events ranging from football matches in stadiums to concerts and festivals in open squares. Comparing our results with the best groundtruth data available, our estimates provide a median error of less than 15% of the actual number of attendees.