Fig. 5
From: Discovering feature relevancy and dependency by kernel-guided probabilistic model-building evolution
Bubble frequency plot of epistasis discovery with Kiedra (blue) and ReliefF (green). Plots are arranged by epistatic problem in rows, i.e. number of relevant SNPs, and noise pollution in columns, i.e. number of irrelevant features: 5 (left) or 10 (right). One hundred datasets per problem were analysed; a bubble in any plot depicts the relative frequency of correct discovered (#relevant, #noisy) features corresponding to its coordinate (abscissas are #relevant). Largest bubbles are the most frequent result found by Kiedra in each problem; the slices shown in such bubbles illustrate in proportion (hits) how often ReliefF agreed with that finding