This tool uses a machine learning model trained on 3,091 real displacement events recorded across Somalia. It predicts whether an incoming event is likely to displace more than 50 people — giving field officers and NGO coordinators an early window to pre-position supplies and staff.
Fill in the four fields on the right with what you know about the event. The model will return a severity label, a confidence score, and a recommended response action.
Random Forest was selected as the deployed model because it achieves the best balance of recall and precision. High recall means it catches more large events — critical in humanitarian response where missing a crisis is more costly than a false alarm.
Fill in the four fields above and click Run Prediction to receive a severity assessment.
—
—
The model looks at the cause, region, month, and duration of a displacement event and returns one of two labels. Here is what each means in practice.
The model predicts fewer than 50 people will be displaced. Standard monitoring is sufficient. No immediate large-scale resource deployment is needed.
The model predicts more than 50 people will be displaced. This is the threshold where humanitarian organizations typically activate emergency response protocols.
The confidence score shows how certain the model is. A 90% confidence on a Large Event means the model has seen many similar past events that turned large. A 55% confidence means the situation is borderline and warrants close monitoring regardless of the label.
Based on 3,091 recorded events. Regions with more flood activity tend to produce larger displacement events on average.