V.9:4 (163-167): Using Fuzzy Logic in Expert Systems by James F. Derry

Using Fuzzy Logic in Expert Systems by James F. Derry

Stock market followers assess the market environment with numerous technical indicators. For the most part, these indicators are rarely definitively bullish or bearish. It is rare that an indicator can be considered to have "crisp" values, such as true or false. More accurately, a technical indicator is considered —as the term might imply—to be an indication of a situation, not a black-or-white statement of some fact. Artificial intelligence researchers deal with this lack of a clear-cut conclusion with the concept of fuzzy logic. Fuzzy logic, according to L.A. Zadeh, is using propositional logic that is modified to allow the use of any value in the range of 0 to 1, rather than only the "crisp" binary values of 0 and 1. The result is fuzzy logic, with a value falling within the range of 0 to 1 considered partly true to the degree of that particular value. For example, a price/earnings (P/E) ratio of 20 may be considered high. A P/E of 25 is even higher. Can P/Es of 20 and 25 both be considered high, but 19 not? A decision based on a crisp rule is difficult to develop, but by using fuzzy logic a rating or an indexed value from 0 to 1 can be applied. The determination of this value is called the membership function, or a "Mu" value.

FUZZY RULES WITH FUZZY MANNERS

Thus, the use of fuzzy logic in an expert system requires values that are traditionally crisp, such as price and price/earnings ratio data, to be used in a fuzzy manner with accordant fuzzy rules. In general, expert systems have many rules in a knowledge base, and more than one rule may be applied when the knowledge base is processed. The question arises as to which conclusions will be used. Consider the arbitrary rule base:

...