AKARS-1 : an Automatic Knowledge Acquisition and Refinement System

Expert Systems are definitely one of the major success of Artificial Intelligence. Unfortunately, the knowledge acquisition bottleneck hampers the development of Expert Systems in practical domains. Machine Learning is a promising solution to this problem and this paper discusses an Automatic Knowledge Acquisition and Refinement System(AKARS-1) for knowledge in rule form. Two major components AKARS-1, the learning component (AKA-1) and the Heuristic Refinement System (HERES) are detailed. A prototype of AKA-1 has been implemented and three case studies are used to validate this prototype. A comparison between ID3 and AKA-1 is presented. HERES employs heuristics and statistical methods to guide its search for refinements. These methods are also discussed in the paper.