Automated design of diagnostic systems

摘要

This research effort represents an inquiry into an important problem of automated acquisition, indexing, retrieval, and effective use of knowledge in diagnostic tasks. Its specific goal is to develop an incremental concept formation system which will automate both the design and use of diagnostic knowledge-based systems by a novice. The adopted approach to this problem is based on the modified family resemblance and contrast model theories, as well as a context-sensitive, distributed probabilistic representation of learned concepts. These ideas have been implemented in the INC2 system. The system is evaluated in terms of its prediction accuracy in the domains of breast cancer, primary tumor, and audiology cases.