An automatic methodology for construction of multi-classifier systems based on the combination of selection and fusion

Abstract

We present a methodology for the automatic construction of multi-classifiers systems based on the combination of selection and fusion. The proposed methodology initially finds the optimum number of clusters for training data set and subsequently determines an ensemble for each cluster found. Self-organizing maps were used in the clustering phase, and multilayer perceptrons, in the classification phase. Adaptive differential evolution was used in order to optimize the parameters and performance of the techniques employed in the classification and clustering phases. The proposed methodology, called SFJADE, was applied on data compression of signals generated by artificial nose sensors and a variety of known classification tasks, including cancer, card, diabetes, glass, heart, horse, soybean, and thyroid. The experimental results shown that the SFJADE methodology had a better performance than some methods reported in related literature and significantly outperformed the methods commonly used to construct multi-classifier systems, for instance, bagging, boosting, and random subspaces.