Benchmark study of feature selection strategies for multi-omics data

BMC Bioinformatics

Table 1 The best performing methods (according to the AUC) per setting

nvar	selsep	clivar	Selector	AUC	Brier	accuracy
10	Yes	Yes	mRMR	0.8299	0.1347	0.8217
10	Yes	No	mRMR	0.8266	0.1357	0.8189
10	No	Yes	mRMR	0.8263	0.1323	0.8281
10	No	No	mRMR	0.8247	0.1331	0.8261
100	Yes	Yes	mRMR	0.8405	0.1287	0.8359
100	Yes	No	mRMR	0.8406	0.1286	0.8363
100	No	Yes	mRMR	0.8345	0.1307	0.8311
100	No	No	mRMR	0.8354	0.1307	0.8290
1000	Yes	Yes	mRMR	0.8374	0.1342	0.8196
1000	Yes	No	mRMR	0.8376	0.1339	0.8200
1000	no	yes	mRMR	0.8290	0.1364	0.8171
1000	No	No	mRMR	0.8274	0.1366	0.8172
5000	Yes	Yes	mRMR	0.8264	0.1383	0.8148
5000	Yes	No	mRMR	0.8260	0.1384	0.8128
5000	No	Yes	mRMR	0.8227	0.1401	0.8111
5000	No	No	mRMR	0.8215	0.1402	0.8107
-	Yes	Yes	Lasso	0.8387	0.1335	0.8219
-	Yes	No	Lasso	0.8413	0.1330	0.8219
-	No	Yes	Lasso	0.8190	0.1374	0.8205
-	No	No	Lasso	0.8185	0.1386	0.8213

The values of the performance metrics were obtained by averaging over the cross-validation repetitions and datasets; ‘nvar’ denotes the number of selected features, ‘selsep’ whether the features were selected separately by data type, and ‘clivar’ whether clinical variables were included or not

ISSN: 1471-2105