查看文章

Uncertainties are unavoidable in real world applications. Hence, any methodology that successfully makes the transition from theory to practice should be equipped with a certain degree of robustness against imprecise data. In the field of classification, an outstanding algorithm needs to provide classification accuracy and be robust against uncertainties that exist during the design and implementation phases. A classifier may be deemed to be robust if it is less sensitive to data variations, and is able to handle insufficient training data scenario [61]. Likewise, in [62] the robustness of the classifier is associated with immunity against missing values in training data and test data. Nanopoulos et al.[63] considered a classifier to be robust if it is able to handle noise which can disrupt the learning process. In other words, a classifier is said to be robust if it is capable of coping well with uncertainties (arising from deficiencies in the available information caused by incomplete, imprecise, ill-defined, not fully reliable, vague, and contradictory information) in various stages of classifier design.

The popularity of fuzzy pattern classification stems from the fact that a FRBC provides a framework to incorporate both subjective (ie, expert opinion) and objective (ie, design samples where the knowledge can be extracted) information, hence it may be able to outperform other classifiers [17]. It is possible to integrate this valuable knowledge into the fuzzy logic system due to the system’s similar reasoning style to the human being. However, an ordinary (type-1) fuzzy set does not capture uncertainty in all of its manifestations, particularly when it arises from vagueness in the …