Constructing fuzzy ensembles for pattern classification problems

This paper examines the performance of fuzzy classifier ensembles. In our fuzzy ensembles, there are multiple fuzzy rule-based classification systems and a single credit assignment system. The credit assignment system is generated from the classification results of individual fuzzy rule-based classification systems. Thus, the credit assignment system plays a role in mapping input space to a fuzzy rule-based classification system. Then the fuzzy rule-based classification system that is selected by the credit assignment system maps pattern space to class. In computer simulations in this paper, we show how our fuzzy classifier ensemble works on a simple two-dimensional pattern classification problem. We also show the classification performance on four real-world pattern classification problems: iris, appendicitis, cancer, and wine data sets. From the results of the computer simulations, we illustrate the low similarity between the fuzzy rule-based classification systems in our fuzzy classifier ensemble lead to high classification performance.