Fig. 3

ROC-AUC, Sensitivity, and Specificity scores using an increasing number of the commonly selected SNPs from RFECV based on their mean feature importance scores for prediction of RA. Each of the commonly selected SNPs from RFECV were gradually included based their feature importance scores (from highest to lowest) in the evaluation of using fivefold cross-validation of training set, unseen test set 1, unseen test set 2, and unseen test set 3. Evaluation scores (ROC-AUC, Sensitivity, Specificity) were plotted against the number of selected SNPs (# SNPs) included in the prediction model. Dotted vertical line in each plot represented the determined optimal number of SNPs for a good evaluation score