Fig. 2: Predictive performance improvements across models.

Changes in average precision (AP) and area under the receiver operating curve (AUROC) scores compared to the Baseline model for Violence, Disasters/Hazards, and Full Distress models (a), and against the Full Distress model for Violence and Disasters/Hazards (b). Shaded areas indicate 95 percent confidence intervals from 250 training and testing repetitions. In (b), negative values signify performance improvements of the Full Distress model relative to the comparison models.