In the previous chapter, we discuss extraction of the three types of fuzzy rules: those with pyramidal membership functions, those with polyhedral regions, and those with ellipsoidal regions using the data included in the associated clusters. Since these fuzzy rules are generated without considering the overlap between classes, their classification performance may not be good. Therefore to improve their classification performance, in this chapter we discuss tuning the slopes and the locations of the membership functions. The direct methods directly maximize the recognition rate of the training data by counting the net increase in the recognition rate when the slope or location of the membership function is changed [56, 72]. The indirect methods maximize the continuous objective function that leads to improving the recognition rate. Finally, we evaluate the performance of the tuning methods for the benchmark data.
Unable to display preview. Download preview PDF.