RNA profiling of blood platelets noninvasively differentiates colorectal cancer from healthy donors and noncancerous intestinal diseases: a retrospective cohort study

Table 1 Classification metrics of SVM

Performance metrics	The classification metrics of SVM across three data sets
Performance metrics	Training set (LOOCV, n = 202)	Internal validation set (n = 120)	External validation set (n = 101)
Accuracy (95% CI)	0.876 (0.823–0.918)	0.875 (0.802–0.928)	0.861 (0.778–0.922)
Sensitivity (95% CI)	0.975 (0.913–0.997)	0.885 (0.766–0.956)	0.761 (0.612–0.874)
Specificity (95% CI)	0.811 (0.731–0.877)	0.868 (0.764–0.938)	0.945 (0.849–0.989)
Positive predicted value	0.772 (0.678–0.850)	0.836 (0.712–0.922)	0.921 (0.786–0.983)
Negative predicted value	0.980 (0.930–0.998)	0.908 (0.810–0.965)	0.825 (0.709–0.909)
Kappa^a	0.752	0.747	0.717
F₁^a	0.862	0.860	0.833

^aKappa measured the agreement between the predicted classification with true labels. F₁ was the harmonic average of precision (positive predicted value) and recall rates (sensitivity)

ISSN: 1756-994X