Figure 2.

Predictive performances of the classifier in the test set. (a) This confusion matrix showed that, among 17 healthy volunteers and 19 patients with CFS in the test set, 27/36 subjects were correctly predicted, resulting in 75% actual accuracy score. (b) The precision-recall matrix showed the average precision (AP) of the classifier was 78%. (c) The area under the receiver operating characteristic curve (ROC AUC) of the classifier was 73%. The (a–c) are three types of measurements to evaluate predictive performances of the classifier. Then we performed half-split 5,000 times permutation tests to determine the significant level of the classifier, and results (d) showed that the balanced accuracy score was 70% and the p value was 0.023, suggesting that the classifier was statistically significant to recognize CFS.