Future Displacement Forecasts

The Foresight model is based on a theoretical framework that focuses on the root causes or macro-level drivers of displacement. The dimensions and associated indicators have been grouped into five categories: 1. Economy: Covers the economic well-being and equality in a given country 2. Security: Covers the level of violence, different types of violence and fatalities 3. Political/Governance: Covers aspects related to the legitimacy of the state, public service provisions and human rights 4. Environment: Covers aspects related to climate disasters, access to water, agricultural stress and food security 5. Societal: Covers aspects related to marginalised groups, urbanisation, size and composition

The data is all derived from open source. The main data sources are the World Bank development indicators, the Armed Conflict Location & Event Data Project (ACLED), the Uppsala Conflict Data Program (UCDP), EM-DAT, UN agencies (UNHCR, the World Food Programme, The Food and Agriculture Organization), Internal Displacement Monitoring Center (IDMC), etc. In total, the system aggregates data from 18 sources, and contains 148 indicators.

The machine learning model employed is an Ensemble. An Ensemble model works by leveraging several constituent models to generate independent forecasts that are then aggregated. Here we employ two gradient-boosted trees to generate the point forecasts. The model hyperparameters were determined by means of a grid search. Each year-ahead forecast has a separate model. In other words, we train a set of Ensemble models for y(t + h) = f(x(t)), where h = 0, 1, 2, 3. The associated confidence intervals were generated by empirical bootstrap method, where the source error distributions were generated on a retrospective analysis. Model training data was limited to data from 1995 onwards.

Overall, the average margin of error of the 188 forecasts made so far is 19%. 50% of the forecasts have a margin of error below 10% and almost 2/3 of the forecasts are less than 15% off the actual displacement.

The model tends to be conservative. Of the current +210 forecasts derived from the model, approximately 60% underestimate the level of displacement for the coming year. The forecasts are based solely on data and developments up until the previous year (i.e. 2022). As such, recent developments are not taken into account. As an example, the Ukraine war that erupted in 2022 was not taken into account in the preliminary or final displacement forecasts made for Ukraine for 2022 and 2023.

Because the model is built around national-level indicators, it does not perform as well in cases where conflict and displacement are largely regionally confined.

Given the methodology of building on historical trends and patterns, the model generally does not tend to capture unprecedented developments.