Abstract

Humanitarian responders in active-conflict areas require timely indications of when civilian needs will arise, but unclassified information, situation reports, and conflict event streams lack clarity. This study provides a replicable OSINT pipeline for transforming these data into a cluster-resolved Humanitarian Need Pressure Index (HNPI) and short-term alerting system. The process is comprised of open data collection, clear need categorization, weighting by severity, geoparsing, aggregation by week per region, lagged regression, and ethical review. In a Ukraine pilot from 2 September 2024 - 31 August 2025, 17,511 news articles and 37,978 geolocated conflicts generated prominent clusters for health, shelter, and displacement needs, which were most pronounced in Donetsk, Kyiv and Kharkiv regions. A ridge regression model of the HNPI one to four weeks ahead showed a held-out MAE of nearly 5.7 points and outperformed persistence, and the surge alert yielded an AUC near 0.91. As the target is the HNPI, the results reflect inherent predictability not independent humanitarian need validation.

Statistics