An expert system to achieve fuzzy interpretations of validation data

Validation of an intelligent system consists, basically, of ensuring that the solutions proposed by the system inspire a level of confidence similar to that inspired by human experts in the application domain. We focus on a validation against a group of experts because represents the ideal balance between the simplicity of using a single expert and the complexity of arriving at a consensus between a number of domain experts. In this paper, a system for automated interpretation of validation data is described. This system is divided into two modules: an algorithmic module functioning with unprocessed data for the statistical measures, which produces output in the form of high-level information that takes working context into account; and a heuristic module, which processes this high-level information to produce the final interpretation. Previous works in the subject highlighted the importance of dealing with the imprecision inherent to the semantic labels fed into the heuristic module by the algorithmic module. One of the typical and most effective ways for handling this kind of imprecision is to use fuzzy logic.