The prediction accuracy was further evaluated using ROC because it is critical for capturing the true positive values in drought impact prediction, namely to predict drought impact when the drought impact occurs (Table 2 and Figure 7). The decision-tree-based models, XGB and RF, showed better performance in all four cases. The AUCs of XGB were the highest, with an average of 0.99 over the four regions, while RF showed the second-best performance, with an average AUC of 0.96. In Gangwon, LL and SVM also showed reasonable performance, with AUC values of 0.92 and 0.86, respectively. In these regions, other than Gangwon, the performances of LL and SVM were worse than those of XGB and RF.