Skip to main content
. 2022 Nov 25;78(Pt 12):1412–1427. doi: 10.1107/S2059798322010415

Figure 4.

Figure 4

Depiction of the results of the feature-permutation analysis for the three classifiers being assessed: support-vector machines (SVM, red), random forest (yellow) and k-nearest neighbours (KNN, turquoise). Each violin depicts the distribution of values observed for the Accuracy decrease recorded after each independent feature permutation. Black bars inside the violins depict the interquartile range, with a white dot showing the median and black whiskers the maximum and minimum quartiles. Note that ‘Accuracy decrease’ here refers to the Accuracy achieved by the trained classifiers with the hold-out test set after feature permutation, while the ‘Accuracy’ shown on the vertical axis refers to the metric used as a predictive feature.