Skip to main content

Table 2 Zero-one and Balanced error rates for the remote homology detection problem optimized for the balanced loss function.

From: Building multiclass classifiers for remote homology detection and fold recognition

   sf95 sf40
   ZE BE ZE BE
MaxClassifier   14.7 30.0 21.0 29.7
Direct K-way Classifiers   11.5 23.1 10.9 13.0
Two-Level Approaches Without Hierarchy Information
  Scaling 9.3 16.1 10.9 13.9
Ranking Perceptron Scale & Shift 10.1 19.5 12.1 15.8
  Crammer Singer 14.7 28.9 17.6 24.1
  Scaling 9.0 15.9 11.8 15.7
SVM-Struct Scale & Shift 10.7 19.9 12.1 15.1
  Crammer Singer 11.6 19.4 13.0 16.3
Two-Level Approaches With Fold-level Nodes
  Scaling 11.2 19.6 14.7 21.4
SVM-Struct Scale & Shift 10.1 19.3 12.1 16.9
  Crammer Singer 14.7 26.0 13.0 18.2
Two-Level Approaches With Class-level and Fold-level Nodes
  Scaling 11.2 20.2 13.0 18.8
SVM-Struct Scale & Shift 13.5 24.7 12.1 16.8
  Crammer Singer 14.7 26.1 13.0 17.5
  1. ZE and BE denote the zero-one error and balanced error percent rates respectively. The results were obtained by optimizing the balanced loss function.