Generating Membership Functions for a Noise Annoyance Model from Experimental Data

The success of fuzzy expert systems could be mainly attributed to the inclusion of linguistic terms into their reasoning scheme. This allows reasoning about complex issues within a certain (tolerated) degree of imprecision. Hence, an important issue in the development of such systems is the choice of the membership functions that model the linguistic terms involved in the application. In this chapter we will describe several methods for the construction of these membership functions (which represent information) from measurements obtained in psycholinguistic experiments. Special attention will be paid to the inclusive and the non-inclusive interpretation of linguistic terms. Secondly, these techniques are applied to data gathered in an International Annoyance Scaling Study, where the relationship between more than 20 different linguistic terms and their corresponding noise annoyance level was under survey.