Inhaltszusammenfassung

Nested dichotomies are binary trees that recursively split a set of classes from a multi-class classification problem into smaller and smaller subsets. In statistics, they are a standard technique for tackling polytomous (i.e. multi-class) classification problems with logistic regression by fitting binary logistic regression models to the individual dichotomous (i.e. two-class) problems at the tree's internal nodes. However, there are usually many possible nested dichotomies for a given set o...Nested dichotomies are binary trees that recursively split a set of classes from a multi-class classification problem into smaller and smaller subsets. In statistics, they are a standard technique for tackling polytomous (i.e. multi-class) classification problems with logistic regression by fitting binary logistic regression models to the individual dichotomous (i.e. two-class) problems at the tree's internal nodes. However, there are usually many possible nested dichotomies for a given set of classes and this technique is only recommended if a particular choice of dichotomies is substantively compelling [Fox, 1997, page 472] based on domain knowledge. In many practical applications there is no a priori reason to prefer one particular grouping of classes over another one. In that case it makes sense to assume that every nested dichotomy is equally likely and to use an ensemble of nested dichotomies for prediction. This is the approach we propose in this paper. Using C4.5 and logistic regression as base learners we show that ensembles of nested dichotomies produce more accurate classifications than applying C4.5 and logistic regression directly to multi-class problems. We also show that they produce more accurate classifiers than pairwise classification if both methods are used with C4.5, and comparable results if they are used with logistic regression. Compared to standard error-correcting output codes, nested dichotomies are preferable if logistic regression is used, and comparable if C4.5 is employed. An additional benefit of nested dichotomies is that they generate class probability estimates. Consequently they appear to be a good general-purpose method for upgrading binary classifiers (that can generate class probability estimates) to multi-class problems.» weiterlesen» einklappen

Autoren

Klassifikation

DFG Fachgebiet:
4.43 - Informatik

DDC Sachgruppe:
Informatik