A modified fuzzy inference system for pattern classification

The use of fuzzy inferencing systems in pattern classifiers and expert systems is now more popular as the linguistic descriptions of inputs helps to deal with input uncertainty. A problem with these systems, however, is that outputs are monotonic and can only add to an output when extra information is acquired. This paper looks at a possible solution to the problem, which involves the inhibition of some rules´ output by other rules making the classification of certain difficult patterns easier. This inhibition is achieved by redefining the consequent NOT function, such modification enables rules to describe holes in the data. Several methods of incorporation are proposed, followed by some areas of suggested usage.