A fine-tuned random forest model categorized global disaster events into impact classes using spatiotemporal and contextual data available at initial reporting.