Building multiclass classifiers for remote homology detection and fold recognition

BMC Bioinformatics

Table 4 Zero-one and Balanced error rates for the fold recognition problem optimized for the balanced loss function.

		fd25		fd40
		ZE	BE	ZE	BE
MaxClassifier		42.0	60.3	44.4	64.6
Direct K-way Classifiers		38.4	52.3	40.4	56.9
	Scaling	39.5	48.7	32.5	48.0
Ranking Perceptron	Scale & Shift	38.8	51.0	29.0	43.0
	Crammer Singer	37.7	49.6	36.0	49.6
	Scaling	39.9	52.7	30.8	46.6
SVM-Struct	Scale & Shift	39.9	52.5	28.1	42.8
	Crammer Singer	41.3	50.5	31.1	43.3
Two-Level Approaches With Class-level Nodes
	Scaling	39.2	52.4	29.9	45.0
SVM-Struct	Scale & Shift	38.1	51.6	29.0	41.7
	Crammer Singer	41.7	50.9	29.9	41.7
Two-Level Approaches With Superfamily-level Nodes
	Scaling	40.2	52.6	30.5	44.5
SVM-Struct	Scale & Shift	40.6	52.7	29.3	42.8
	Crammer Singer	38.8	48.8	31.0	44.9
Two-Level Approaches With Superfamily-level and Class-level Nodes
	Scaling	41.0	50.9	33.7	44.6
SVM-Struct	Scale & Shift	39.5	51.5	29.3	42.3
	Crammer Singer	40.2	51.9	30.2	42.4

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

ISSN: 1471-2105