Fig. 3
From: On the performance of pre-microRNA detection algorithms
Model training workflow. Filtered human miRNA hairpins from miRBase served as positive data and pseudo hairpins for negative data. Each data set is randomly sampled individually; 70% of positive data and the same number of negative examples were used during 1000-fold Monte Carlo cross validation (MCCV) 30. The remaining 30% of positive data and the same number of negative examples are used for testing the model. In the end, the best models for naïve Bayes and decision tree were stored for prediction in PMML format while SVM performance was not stored as a PMML model due to limitations of the available SVM implementation
