Fig 4. Performance of bagged decision trees (BGDTs) classifying exemplars of courtship and non-courtship motifs.

Plotted are confusion matrices that indicate the actual (y-axis) categories (courtship vs. non-courtship) and predicted categories generated by the BGDTs (x-axis) for (A) an individual classifier (male ‘prpred’; 81.7% overall accuracy) and (B) the combined classifier (82.4% overall accuracy). Classification accuracy of the individual or combined classifiers did not significantly correlate with the sample size of the dataset (S1 Fig). (C) Data visualization of courtship and non-courtship motifs using t-distributed stochastic neighbor embedding (tSNE), a non-linear technique for dimensionality reduction (different symbols within each colored cluster; x is a courtship motif, o is a non-courtship motif). Shown are example spectrograms of non-courtship motifs from three different birds.