Harnessing AI for Humanitarian Planning
Primary and secondary roads in Kenya, filtered by predicted riot risk.
Conflict Forecast App was developed as part of the WFP Innovation Accelerator in collaboration with INNODEMS. This web application helps identify potential 'access disruption' hotspots in conflict-affected regions. The app provides decision-makers with data-driven insights, helping them understand where critical infrastructure might be vulnerable to conflict risks, with potential implications for communities and humanitarian efforts.
How it Works?
We provide sub-national forecasts of conflict intensity at the admin2 level with different forecast horizons. Currently, the forecasts displayed in the Conflict Forecast App are for three months ahead.
Although we are continuously experimenting with different models and ensembles, we typically rely on tree-based ensemble models and implement a rolling forecast methodology. We train models using global data from 2012 onward. With the resulting model, we generate out-of-sample predictions on a rolling basis from 2020 onward.
After obtaining predictions, we overlay them with other data layers to provide early warnings of potential disruptions to infrastructure due to riots or other events that may lead to fatalities.
Key Features
-
Three-Month Ahead Conflict Forecasts
- Predicts the probability of conflict events (e.g., riots) and conflict intensity (measured by fatalities) at the subnational level (admin2) within a three month horizon.
-
Integrated Data Layers
- Overlays conflict predictions with infrastructure and socio-economic datasets to support contingency planning, early procurement, and prepositioning of goods.
-
Interactive Visualization & Alerts
- Provides map-based insights and alert tables, allowing users to filter, sort, and track predicted conflict risks and month-to-month changes.
Real-World Impact
We are currently piloting the Conflict Forecast App with key partners and signed an MoU with the Kenya Red Cross
to test the solution in real-world scenarios. These partnerships will ensure that our tools are used complementary to existing expertise on the ground and provide maximum benefit to our partners.
Learn More
Our work is part of the 6-month Stability & Peace Accelerator.
Check out our pitch here.
For collaboration opportunities or further details, get in touch with us. Together, we can leverage data to create meaningful change in crisis-prone regions.
We provide sub-national forecasts of conflict intensity at the admin2 level with different forecast horizons. Currently, the forecasts displayed in the Conflict Forecast App are for three months ahead.
Although we are continuously experimenting with different models and ensembles, we typically rely on tree-based ensemble models and implement a rolling forecast methodology. We train models using global data from 2012 onward. With the resulting model, we generate out-of-sample predictions on a rolling basis from 2020 onward.
After obtaining predictions, we overlay them with other data layers to provide early warnings of potential disruptions to infrastructure due to riots or other events that may lead to fatalities.
Key Features
-
Three-Month Ahead Conflict Forecasts
- Predicts the probability of conflict events (e.g., riots) and conflict intensity (measured by fatalities) at the subnational level (admin2) within a three month horizon.
-
Integrated Data Layers
- Overlays conflict predictions with infrastructure and socio-economic datasets to support contingency planning, early procurement, and prepositioning of goods.
-
Interactive Visualization & Alerts
- Provides map-based insights and alert tables, allowing users to filter, sort, and track predicted conflict risks and month-to-month changes.
Real-World Impact
We are currently piloting the Conflict Forecast App with key partners and signed an MoU with the Kenya Red Cross
to test the solution in real-world scenarios. These partnerships will ensure that our tools are used complementary to existing expertise on the ground and provide maximum benefit to our partners.
Learn More
Our work is part of the 6-month Stability & Peace Accelerator.
Check out our pitch here.
For collaboration opportunities or further details, get in touch with us. Together, we can leverage data to create meaningful change in crisis-prone regions.
-
Three-Month Ahead Conflict Forecasts
- Predicts the probability of conflict events (e.g., riots) and conflict intensity (measured by fatalities) at the subnational level (admin2) within a three month horizon.
-
Integrated Data Layers
- Overlays conflict predictions with infrastructure and socio-economic datasets to support contingency planning, early procurement, and prepositioning of goods.
-
Interactive Visualization & Alerts
- Provides map-based insights and alert tables, allowing users to filter, sort, and track predicted conflict risks and month-to-month changes.