Connecting...

Predict Protest Repression Level

Pro-Test uses an ensemble of machine learning models trained on historical protest data to predict the likely severity of repression — from no response to lethal force.

How It Works

Step 1

Select Location

Choose the country (Iraq, Lebanon, or Egypt) and specific governorate where the protest will occur.

Step 2

Define Characteristics

Specify the location type, demand type, and primary tactic used by protesters.

Step 3

Estimate Participants

Enter the expected number of participants. Larger protests may have different outcome patterns.

Step 4

Get Predictions

Click 'Predict Outcomes' to receive a repression level prediction (0–5) with a full probability distribution.

Repression Level Scale

Level 0 — None

No known coercion or security presence at the event.

low

Level 1 — Presence

Security forces or repressive groups present at the event.

low

Level 2 — Escalated

Army deployed or participants summoned to a security facility.

medium

Level 3 — Force Used

Physical harassment, arrests, detentions, or militia involvement.

medium

Level 4 — Injuries

Injuries inflicted on protesters by security forces.

high

Level 5 — Lethal

Deaths inflicted. Highest severity repression.

critical

Important Considerations

  • Research Tool: This system is designed for research and risk assessment. It should not be the sole basis for safety decisions.
  • Historical Data: Predictions are based on events from 2017-2022 and may not reflect current political conditions.
  • Regional Scope: Only protests in Iraq, Lebanon, and Egypt are supported. Other regions require additional data.
  • Probabilistic: The output is a predicted severity level with a probability distribution. The actual level may differ.

Frequently Asked Questions

How accurate are the predictions?

The ensemble model achieves high accuracy on historical data for the repression level scale (0–5). However, predictions are based on patterns from 2017-2022 and may not account for recent changes in political dynamics.

What data was used to train the model?

The model was trained on over 13,000 protest events from Iraq, Lebanon, and Egypt, documented between 2017 and 2022. Data includes protest characteristics and documented security responses.

Can I use this for other countries?

Currently, the model only supports Iraq, Lebanon, and Egypt. Predictions for other regions would require additional training data and model validation.

How should I interpret the probabilities?

The model returns a predicted repression level (0–5) and a probability for each level. The highlighted level has the highest probability based on similar historical events, but other levels are possible.

Is this data updated in real-time?

No, the model is trained on historical data. It does not incorporate real-time information about ongoing events or recent political developments.