Figure 2:

The framework of feature ranking process. A) Bootstrapping samples and features. It includes Steps 1 and 2. V_m: the m-th bootstrap feature index; B_m: the m-th bootstrap sample index; $B_m^c$: the m-th bootstrap out-of-bag sample index. B) Pairing data, training the reference model, permuting and recording the decrease in classification performance. This includes Steps 3–6. C) Ranking features based on how often the baseline classification accuracy is reduced when permuted. This includes Steps 7 and 8.