Equivalence and Confidence of Classifiers: Selected Approaches using ROC Curves and Bootstrap

Abstract

The thesis deals with evaluation of classifiers in two types of situations. In supervised learning, we characterize the classifier by ROC curve and we find for two parametric models (binormal and biexponential) tests of equivalence of couple of classifiers and we investigate the properties of the tests.

In the unsupervised learning, we focuse on application of cluster analysis in fylogeny and using bootstrap we estimate the confidence levels of the branches of evolutionary trees.

Keywords

• Equivalence
• Confidence
• Classifiers
• Selected
• Approaches
• using
• Curves
• Bootstrap