MiniBoosting Decision Trees

DMCA

MiniBoosting Decision Trees (1999)

Cached

Download Links

Other Repositories/Bibliography

DBLP

by J. R. Quinlan

Citations:

4361 - 4 self

BibTeX

@MISC{Quinlan99miniboostingdecision,
    author = {J. R. Quinlan},
    title = {MiniBoosting Decision Trees},
    year = {1999}
}

OpenURL

Abstract

Boosting, introduced by Freund and Schapire, is a method for generating an ensemble of classifiers by successive reweightings of the training cases. We study boosting in the context of small ensembles of decision trees, showing that the reweighting procedure can be improved and that the resulting ensemble can be represented by a single decision tree. This tree is large, attesting to the complexity of boosted classifiers, but simplifications that do not affect its performance on the training data destroy the benefits of boosting. 1. Introduction Learning to predict the categories to which objects belong by analysing a collection of training cases is one of the most studied areas of Machine Learning. Well-tried methods for constructing a classifier or function that maps from object descriptions to class names include instance-based approaches (Aha, Kibler, & Albert, 1991), neural networks (McClelland & Rumelhart, 1988), decision trees (Breiman, Friedman, Olshen, & Stone, 1984; Quinlan, ...

Keyphrases

decision tree training case neural network boosted classifier training data studied area successive reweightings instance-based approach machine learning small ensemble object description single decision tree reweighting procedure mcclelland rumelhart class name well-tried method

DMCA