Skip to main content

Table 1 Comparison of the characteristics of the classifiers used in this study. The results are derived from the data in Figures 2-4.

From: Boosting accuracy of automated classification of fluorescence microscope images for location proteomics

Classifier Ability to generate nonlinear decision boundary Ability to learn well from limited training data Insensitivity to outliers in training data Insensitivity to uninformative features Log information content* (2D/3D)
Neural Networks Low High Medium Medium 10.0/10.0
Exponential-rbf-kernel SVM High High High Low 14.2/13.9
AdaBoost Medium Med High High 13.5/13.4
Bagging Medium Low High High 14.8/12.0
Mixtures-of-Experts Medium Low High High 13.5/13.5
Majority-voting Ensemble Medium High High High 14.7/14.6
  1. * The natural logarithmic of the information content was calculated as described in the Methods section for the feature set SLF13 (2D) and SLF10 (3D). The classifiers were configured as detailed in Table 2.