Evolutionary Boosting for Improving Accuracy of Classifier Ensembles

K. Matsui and H. Sato (Japan)


Boosting, Ensemble Learning, Genetic Algorithm, Index Selection


We propose a new boosting method for improving accu racy of classifier ensembles using an evolutionary tech nique. This method is called EBIS (Evolutionary Boost ing with Index Selection). EBIS is an ensemble-learning method which involves two important concepts; AdaBoost and Evolutionary Index Selection (EIS). The former is one of the most popular boosting methods in ensemble learn ing. The latter is an index-selection method using Ge netic Algorithm for pattern-classification tasks. In EIS, an entropy-based criterion, VQCCE (Vector-quantized Condi tional Class Entropy), is used for the evaluation of index combinations. Our EBIS generates an accurate classifier ensemble by using EIS at each round of AdaBoost. We apply our EBIS to some test-problems and discuss the ef fectiveness of our method.

Important Links:

Go Back