Hybrid Expert System

This article deals with hybrid expert system that has knowledge base realized through a hierarchical structure of artificial neural networks (NN). The decision tree is built by C4.5 algorithm at first. In the next step the nods of the tree are replaced by NN. They are trained to split the data in the same way as the nods. So the problem is separated into partial sub-problems that are solved by individual NN. At the end an expert is requested to change the structure of particular NN according to his knowledge and experiences. Each NN solves a partial sub-problem what decreases demands upon the capability of an expert and accelerates the time needed to harmonize the knowledge base. Unlike the traditional expert system, the output of this architecture is not classification, however we receive a list of hypotheses evaluated by certain value of the hypothesis trust. Key-Words: expert system, connectionist expert system, knowledge base, neural network, decision tree