Solutions to the class imbalance problem. The figure shows, for three classifiers, the overall (PA) and class specific predictive accuracies (PA1 and PA2), varying the proportion of Class 1 samples in the training set (). The simulation settings are the same as those from the left panels of Fig. 3 (p = 1000). The results are given for the situations when no class imbalance correction is applied (No correction, first column), or when over-sampling (second column), down-sizing (third column) or multiple down-sizing (MultDS, forth column) are used.