📱 Protests Leave a Digital Trace
When tens of thousands of people gather in the streets, they don’t just fill public squares — they reshape the mobile network.
In this work :contentReference[oaicite:1]{index=1}, we show that large public protests leave a distinct and measurable footprint in mobile network traffic. By analyzing aggregate data from a nationwide operator during the 2023 French pension reform protests, we demonstrate that it is possible to:
- Detect protests from network-wide traffic fluctuations
- Reconstruct their spatiotemporal evolution
- Estimate time-varying attendance
- Do so using privacy-preserving aggregated measurements
Importantly, our approach does not rely on social media content, user-level tracking, or individual identification.
🔎 What Makes Protest Traffic Unique?
Using 5-minute traffic aggregates across thousands of cellular carriers, we first compare protest days with baseline days.
We find two strong macroscopic effects:
1️⃣ Overall Traffic Surges Along the Route
Carriers near protest routes exhibit statistically significant traffic deviations compared to normal days.
These deviations evolve over time, naturally following the physical progression of the march.
2️⃣ Application Usage Shifts Dramatically
Protests reshape how people use their phones.
Applications that increase in relative importance:
- Twitter/X
- Google Maps
- News platforms
Applications that decrease in relative importance:
- Netflix
- Spotify
- Video streaming services
Rather than looking at raw traffic volumes (which vary widely), we introduce a normalized share-based metric that captures changes in application importance relative to baseline conditions.
The result: a clear digital signature of protest activity.
🤖 Modeling Protest Detection
The footprint is strong enough to enable automatic detection.
We train an XGBoost classifier using application-level traffic share variations as features. The model predicts, at 5-minute resolution, whether a cellular carrier is affected by protest activity.
Performance:
- F1-score: 0.97
- Precision: 0.99
- Recall: 0.95
To ensure spatial and temporal coherence, we combine classification with density-based spatiotemporal clustering (ST-DBSCAN).
The outcome:
- Accurate reconstruction of protest routes
- Identification of alternative or unofficial paths
- Clear temporal progression from start to end
The method generalizes across cities — models trained in Paris successfully detect protests in Lyon, Toulouse, Nantes, and Bordeaux.
👥 Estimating Protest Attendance from Traffic
Can network traffic estimate crowd size?
We find a strong power-law relationship between:
Peak traffic volume at affected carriers
and
Official attendance estimates
Using this relationship, we reconstruct dynamic attendance curves over time.
Typical pattern:
- Gradual buildup
- Mid-event peak
- Dispersal toward the end
This enables high-resolution post-hoc analysis of:
- Protest intensity
- Participation dynamics
- Spatial concentration shifts
All without tracking any individual.
🔐 Privacy Considerations
A key aspect of this study is that:
- Data are fully aggregated at carrier level
- No user identifiers are used
- No device-level analysis is performed
- Small gatherings are not detectable
The methodology is inherently limited to large-scale events, which prevents individual-level surveillance.
This work demonstrates that mobile network measurements can provide societal-scale insights while preserving privacy.
📌 Why This Matters
This research opens a new direction in:
- Network measurement
- Urban analytics
- Computational social science
It shows that large collective behavior leaves measurable traces in infrastructure systems — and that these traces can be analyzed responsibly.
Beyond protests, the framework could apply to:
- Large festivals
- Sports events
- Emergencies
- Urban crowd management
📖 Reference
André Felipe Zanella et al.
Characterizing, Modeling and Exploiting the Mobile Demand Footprint of Large Public Protests
IMC 2024 :contentReference[oaicite:2]{index=2}