Multiple Species Weighted Voting – A Genetics-Based Machine Learning System

Alexander F. Tulai and Franz Oppacher

Abstract. Multiple Species Weighted Voting (MSWV) is a genetics-based machine learning (GBML) system with relatively few parameters that combines N two-class classifiers into an N-class classifier. MSWV uses two levels of speciation, one manual (a separate species is assigned to each two-class classifier) and one automatic, to reduce the size of the search space and also increase the accuracy of the decision rules discovered. The population size of each species is calculated based on the number of examples in the training set and each species is trained independently until a stopping criterion is met. During testing the algorithm uses a weighted voting system for predicting the class of an instance. MSWV can handle instances with unknown values and post pruning is not required. Using thirty-six real-world learning tasks we show that MSWV significantly outperforms a number of well known classification algorithms.

LNCS 3103, p. 1263 ff.

Full article in PDF

lncs@springer.de
© Springer-Verlag Berlin Heidelberg 2004