Abstract: Classification of biomedical data faces a special challenge because
of the characteristics of the data: too few data examples with too many
features. How to improve the classification performance or the generalization
ability of a classifier in the biomedical domain becomes one of the active
research areas. One approach is to build a fusion model to combine multiple
classifiers together and result in a combined classifier which can achieve a
better performance than any of its composing individual classifiers. In this
paper, we propose a SVM classifier fusion model to combine multiple SVMs by
applying the knowledge of fuzzy logic and genetic algorithms. The fuzzy logic
system (FLS) is constructed based on SVM accuracies and distances of data
examples to SVM hyperplanes in SVM feature spaces. A genetic algorithm (GA) is
used to tune the fuzzy membership functions (MFs) in the FLS and determine the
optimal fuzzy fusion model. We have applied the proposed model to two
biomedical data: colon tumor data and ovarian cancer data. Our experiment shows
that multiple SVM classifiers complement each other well in the proposed fusion
model and the ensemble achieves a better, more robust and more reliable
performance than individual composing SVMs.
Keywords: Fuzzy logic, evolutionary computation, genetic algorithms, bioinformatics, medical informatics, support vector machines, ensembles, classification
