Combinacion de clasificadores basados en mezclas gaussianas (Spanish text).

摘要

Several studies have shown that the combination of unstable classifiers such as decision trees, neural networks and classifiers based on kernel density estimators improve the performance of single classifiers. In this investigation we present some results of the effect of the combination of classifiers based on Gaussian mixtures, by means of the algorithms Bagging and Boosting, applied to eleven real-world datasets. In addition we study the instability of the classifiers and the effect of the feature selection on the error of classification as well as on the performance of Bagging and Boosting. Our results show that the classifier based on Gaussian mixtures is unstable, but its degree of instability depends on the structure of the data set. Also the missclassification rate reduction for Bagging is superior to Boosting, and its performance is more uniform.