Classification process analysis of bioinformatics data with a support vector fuzzy inference system

• Authors: Stergios Papadimitrioy;Konstantinos Terzidis
• Venue: NN'07 Proceedings of the 8th Conference on 8th WSEAS International Conference on Neural Networks - Volume 8
• Year: 2007

Recent complex bioinformatics data sets, such as Microarray and Proteomics data sets, which are characterized by sparsity and high dimensionality, require an analysis, which on the one hand offers a high degree of accuracy, but on the other hand simultaneously provides transparency in the analysis process. Recent Machine learning techniques, like e.g. the Support Vector Machines, own a remarkable generalization ability and are among the first choices to confront such complex data. However, the black-box structure of most machine learning algorithms constitutes a significant drawback. On the other hand, Fuzzy rule based systems form an attractive alternative since they result in linguistically, interpretable rules, but suffer from the problem of overfitting and are sensitive to the curse of dimensionality. In order to merge the advantages of both approaches Support Vector algorithms have been adapted for the identification of a Support Vector Fuzzy Inference (SVFI) system. However, although the high generalization performance of the SVM approach is retained, the SVFI rules usually lack understandability. The paper proposes the derivation of a simpler fuzzy system that approximates the accurate set of rules keeping only the more important aspects of the data. The approximation algorithms either receive an a priori description of a set of fuzzy sets or, especially for the case when interpretable fuzzy sets cannot be prespecified by the experts, an algorithm is presented for building them automatically. After the construction of the interpretable fuzzy partitions, the developed algorithms extract from the SVFI rules a small and consice set of interpretable rules. Finally, the Pseudo-Outer Product (POP) fuzzy rule selection orders the interpretable rules by using a Hebbian like evaluation in order to present the designer with the most capable rules.