Abstract:
|
Adaboost, a typical boosting method for classification, has performed quite well in classification problems. Many researchers have applied different types of randomization techniques to adaboost for further improving the efficiency of classification. However, these methods of randomization seldom aim at the chance mechanism underlying the training data itself, especially at the response level. We propose two different methods to estimate the conditional probabilities of each class label, and then randomize the class labels based on the estimated probabilities in a modified adaboost procedure. The first method uses the proportion of each class obtained from the output of bagging as a possible estimator of the probability of each class label. The second method makes use of over/under-sampled classification to find a reliable interval for the probability of each class label.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.