Skip to main content
. 2021 Jun 17;2(7):100289. doi: 10.1016/j.patter.2021.100289

Figure 4.

Figure 4

AUROC and AUPRC of classifiers based on different word vectorization methods in 30 bootstrapping experiments on cohort 1

(A) AUROC results. The best model in bootstrapping experiments based on AUROC was TF-IDF (mean AUROC (95% CI)): I21, 0.8952 (0.8768–0.9075); I25, 0.9487 (0.9470–0.9514); I27, 0.9537 (0.9505–0.9585); I42, 0.9763 (0.9735–0.9790); I48, 0.9745 (0.9731–0.9762); I50, 0.9543 (0.9522–0.9571); I70, 0.9185 (0.9046–0.9333); I85, 0.9951 (0.9918–0.9981).

(B) AUPRC results. The best model in bootstrapping experiments based on AUPRC was TF-IDF (mean AUPRC (95% CI)): I21, 0.0723 (0.0549–0.0951); I25, 0.6752 (0.6709–0.6830); I27, 0.5370 (0.5189–0.5557); I42, 0.6079 (0.5949–0.6240); I48, 0.7913 (0.7878–0.7948); I50, 0.5028 (0.4888–0.5161); I70, 0.1941 (0.1344–0.2514); I85, 0.1281 (0.0727–0.2108).