Figure 2From: Cost sensitive hierarchical document classification to triage PubMed abstracts for manual curationComparison of Naïve Bayes and SVM algorithms at training Level 0. The performance of the Naïve Bayes and SVM classifiers was evaluated with 10-fold cross-validation. As is shown in the ROC curve, the SVM classifier outperformed the Naïve Bayes classifier on curatability predictions for the cross-validation dataset of 89,884 abstracts. The AUC value for the SVM classifier was 0.899 and the AUC value for the Naïve Bayes classifier was 0.854. At the 5% false negative rate for the curatability decision, the SVM classifier had a true positive rate of 41.4% and the Naïve Bayes classifier had a true positive rate of 33.5%.Back to article page