A Characteristic-Point-Based Fuzzy Inference Classifier by a Closeness Matrix

In this paper, a characteristic-point-based fuzzy inference classifier (CPFIC) is proposed to perform two-class classification. Through fuzzy interpolation, a subset of classified samples can be taken as representatives of all samples. They are called characteristic points (CPs). A closeness matrix representing the closeness of two samples in a same class is proposed in selecting CPs. By solving a number of constrained minimizations, the CPFIC is systematically built. Experiments were conducted on four classification problems with known Bayes errors, two benchmark classification problems, and a real-world application used in our research. The CPFIC performs well in accuracy evaluations in all the seven experiments. The summarizing abilities from the CPs into the linguistic descriptions of the fuzzy rule bases were also demonstrated in these examples