Automated qualitative description of measurements

Measurements are usually thought of as precise numbers resulting from observation of a real-world system by means of special sensors or measuring devices. Humans frequently resort, however, to qualitative descriptions-based on the extent by which measurements agree with various models of relevance to the system being observed-to explain the nature and importance of a particular measurement or that of a set of related observations. We present results of ongoing research on methods for the automatic derivation of qualitative descriptions of complex objects. The ultimate goals of these investigations are the development of a methodology for the qualitative representation of complex objects, the systematic search and retrieval of measurements and objects based on those representations, and the discovery of knowledge based on the study of collections of such qualitative descriptions. Our techniques combine fuzzy logic and evolutionary computation methods to solve optimization problems associated with qualitative description. These methods are noteworthy in that they do not assume prior knowledge of the number of interesting structures, or their extension nor do they require an exhaustive explanation of the object being described. We present results of the application of these methods to the description of financial time series