The identification of inference rules from fuzzy data: The application of a quantitative model of perception

In many fields involving expert or nonexpert decision-making, it is desirable to have a model for the conversion of input information, as it is measured, perceived, and interpreted, into a consequent response, or classification of events in question. However, the available data may be limited to a number of discrete observations, or tests, from which the more general rules or processes need to be discerned. For example, individuals may react to perceived threats (e.g. food or health risks, nuclear power, transport, site selection for hazardous installations) by classifying them according to a number of distinct (though not necessarily independent) attributes before deciding whether or not to take some positive action. Psychometric studies have identified a number of underlying risk attributes such as familiarity, scientific understanding, potential benefits, catastrophe (as against constancy), effects on future generations etc. All of these are ambiguously defined, and are subjective and context-dependent. Alternatively we may wish to model the process by which a certain type of individual (based on selective social qualifiers) makes decisions from a range of alternatives — for example, the range of considerations in making investments or purchases. Here again the evaluation process involves assessing candidates according to a number of nonquantitative or ill-defined attributes (return on the investment, liquidity, duration, the size of investment), some of which may be interdependent. Moreover, similar considerations employed must be made for client profiling (in insurance), credit scoring (in banking), or ability assessments (in personnel management). In all such problems the attributes may not be exchangeable, but combinations of attributes may be used in defining thresholds or standards, violation of which leads to outright rejection. The rule applied is a logical one, though, dealing with rather ill-defined concepts.