Advances in fuzzy rule-based system for pattern classification

Abstract

Pattern classification encompasses a wide range of information processing problems that are of great practical significance, from the classification of handwritten characters, to fault detection in machinery and medical diagnosis. Fuzzy logic system was initially introduced to solve a pattern classification problem because the system has similar reasoning style to human being. One of the main advantages of fuzzy logic is that it enables qualitative domain knowledge about a classification task to be deployed in the algorithmic structure. Despite the popularity of fuzzy logic system in pattern classification, a conventional singleton type-1 fuzzy logic system does not capture uncertainty in all of its manifestations, particularly when it arises from the noisy input and the vagueness in the shape of the membership function. The aim of this study is to seek a better understanding of the properties of extensional fuzzy rule-based classifiers (FRBCs), namely non-singleton FRBC and interval type-2 FRBC. Besides, this research aimed at systemising the learning procedure for fuzzy rule-based classifier.

Non-singleton FRBC was found to have noise suppression capability. Therefore, it can better cope with input that is corrupted with noise. In addition, the analysis demonstrated that non-singleton FRBC is capable of producing variable boundary which may be useful to resolve the overlapping boundary between classes. The significance is that non-singleton FRBC may reduce the complexity of feature extraction by extending the possibility to use the features that are easier to extract but contain more uncertainties. As an extension to type-1 fuzzy classifier, type-2 classifier appears to have better performance and robustness due to its richness of footprint of uncertainty (FOU) in membership function. The proposed FOU design methodology can be useful when one is uncertain about the descriptions for the features (i.e., the membership function). The robustness study and extensive experimental results suggest that the performance of type-2 FRBC is at least comparable, if not better than type-1 counterpart.

Designing and optimising FRBCs are just as important as understanding the properties of different types of fuzzy classifiers. In view of this, an efficient learning algorithm based on support vector machine and fuzzy c-means algorithm was proposed. Not only that the resulting fuzzy classifier has a compact rule base, but it also has good generalisation capability. Besides, the curse of dimensionality which is often faced by FRBCs can be avoided. In the later part of this thesis, it was also shown that the proposed fuzzy rule-based initialisation procedure can enhance the performance of conventional crisp and fuzzy K-Nearest Neighbor (K-NN) when the training data is limited. Moreover, the successful implementation of the FRBC to classify faults in induction motor has provided clear evidence of its practical applicability.

In conclusion, it is foreseeable that FRBCs will continue to play an important role in pattern classification. With the advances in extensional FRBCs, the uncertainties which the conventional classifiers failed to address for, can now be handled more effectively.