Building multiclass classifiers for remote homology detection and fold recognition

BMC Bioinformatics

Table 3 Zero-one and Balanced error rates for the remote homology detection problem optimized for the zero-one loss function.

		sf95		sf40
		ZE	BE	ZE	BE
MaxClassifier		14.7	30.0	21.0	29.7
Direct K-way Classifiers		13.5	24.8	20.5	26.5
Two-Level Approaches Without Hierarchy Information
	Scaling	10.6	18.0	11.7	16.5
Ranking Perceptron	Scale & Shift	13.2	24.5	10.9	13.4
	Crammer Singer	17.0	34.3	14.2	19.4
	Scaling	10.7	18.1	13.4	17.3
SVM-Struct	Scale & Shift	12.4	23.7	13.4	17.3
	Crammer Singer	12.7	25.2	15.5	19.8
Two-Level Approaches With Fold-level Nodes
	Scaling	10.4	18.7	14.7	20.0
SVM-Struct	Scale & Shift	12.4	23.7	14.7	21.4
	Crammer Singer	13.8	25.0	14.7	19.6
Two-Level Approaches With Class-level and Fold-level Nodes
	Scaling	10.9	19.1	12.6	17.7
SVM-Struct	Scale & Shift	11.2	20.9	13.4	17.8
	Crammer Singer	14.1	27.6	12.6	17.1

ZE and BE denote the zero-one error and balanced error percent rates respectively. The results were obtained by optimizing the zero-one loss function.

ISSN: 1471-2105