A comprehensive evaluation of multicategory classification methods for microbiomic data

Microbiome

Table 7 Classification accuracy with feature/ operational taxonomic unit (OTU) selection, measured by relative classifier information (RCI)

Classifier	Best FS Method	CBH	CS	CSS	FS	FSH	BP	PDX	PBS	Averages	P values
SVM, Linear C = 1	SVM-RFE	0.719	0.952	0.691	0.929	0.813	0.334	0.337	0.674	0.681	0.191*
*SVM, Linear optimized C*	*SVM-RFE*	0.852	0.946	0.723	0.971	0.840	0.314	0.325	0.653	0.703	-
SVM, Poly	SVM-RFE	0.845	0.941	0.716	0.969	0.840	0.316	0.323	0.644	0.699	0.369*
SVM, RBF	SVM-RFE	0.813	0.925	0.683	0.972	0.813	0.286	0.290	0.611	0.674	0.089*
KRR, Poly	SVM-RFE	0.759	0.939	0.683	0.931	0.800	0.297	0.290	0.626	0.666	0.061*
KRR, RBF	SVM-RFE	0.807	0.935	0.687	0.944	0.801	0.297	0.316	0.633	0.677	0.097*
KNN, K = 1	RFVS2	0.830	0.779	0.657	0.939	0.736	0.168	0.251	0.510	0.609	0.015
KNN, K = 5	RFVS2	0.774	0.744	0.625	0.884	0.736	0.153	0.224	0.522	0.583	0.008
KNN, optimized K	RFVS2	0.829	0.773	0.652	0.914	0.736	0.179	0.221	0.531	0.604	0.014
PNN	RFVS2	0.726	0.798	0.629	0.907	0.730	0.167	0.227	0.516	0.587	0.012
L2-LR, C = 1	ALL	0.772	0.941	0.670	0.964	0.778	0.161	0.236	0.628	0.644	0.027
L2-LR, optimized C	SVM-RFE	0.780	0.940	0.692	0.837	0.811	0.234	0.257	0.612	0.645	0.034
L1-LR, C = 1	RFVS1	0.742	0.836	0.642	0.934	0.771	0.183	0.213	0.584	0.613	0.011
L1-LR, optimized C	RFVS1	0.786	0.914	0.696	0.985	0.784	0.166	0.238	0.598	0.646	0.033
RF, default	RFVS1	0.840	0.952	0.712	0.982	0.819	0.266	0.213	0.648	0.679	0.179*
RF, optimized	RFVS1	0.842	0.956	0.714	0.994	0.810	0.264	0.216	0.649	0.681	0.196*
BLR, Laplace priors	SVM-RFE	0.822	0.932	0.692	0.982	0.824	0.317	0.318	0.640	0.691	0.313*
BLR, Gaussian priors	RFVS2	0.761	0.855	0.625	0.968	0.770	0.208	0.202	0.570	0.620	0.018

The nominally best performing classifier on average over all datasets is marked with bold, and P values of methods whose performance cannot be deemed statistically worse than the nominally best performing method are marked with “*”. The accuracy of the nominally best performing method for each dataset is underlined.

ISSN: 2049-2618