Skip to main content
. 2022 Mar 31;3(5):100471. doi: 10.1016/j.patter.2022.100471

Table 1.

False discovery rate for the synthetic datasets

Synthetic 1 [60 × 30, C2] Synthetic 2 [1,000 × 100, C2] Synthetic 3 [1,000 × 500, C10] Synthetic 4 [100 × 500, C8]
GSO 0 0 0 0.10
mRMR Peng 0.67 0 0.30 0.90
mRMR Spearman 0.33 0 0.20 0.60
Information gain 0.67 0 0.50 1
RELIEF 0.33 0.38 0.30 0.70
CFS 0.67 0.13 0.50 1
CBF 0.67 0.13 0.50 1
SIMBA 0.33 0.50 1 0.80
LOGO 0.67 0.25 0.30 0.70
L1-LSMI 0.33 0.13 0.60 0.90
IAMB 0 0 0 0.10
HITON 0.67 0 0.90 1
JMI 0.33 0 0.30 0.90
DISR 0.33 0 0.30 0.90
QPFS 0.67 0 0.30 0.90
CMIM 0.66 1 0.40 0.90
CIFE 0.33 1 0.50 1
MIQ 0.67 0.13 0.40 0.80
SPECCMI 0.67 1 0.30 1
RRCT 0 0 0 0.10

The design matrices are summarized in the form N×M [number of samples × number of features], and the following term indicates the problem and number of classes (e.g., C2 indicates that this is a classification problem with two classes). The presented results are the FDR scores for the number of true features in each of the datasets (see Figure 1 and also the description of the synthetic datasets for details).