CONSULTANT-2 - PRE-PROCESSING AND POST-PROCESSING OF MACHINE LEARNINGAPPLICATIONS

The knowledge acquisition bottleneck in the development of large knowl edge-based applications has not yet been resolved. One approach which has been advocated is the systematic use of Machine Learning (ML) tech niques. However, ML technology poses difficulties to domain experts an d knowledge engineers who are not familiar with it. This paper discuss es Consultant-2, a system which makes a first step towards providing s ystem support for a ''pre- and post-processing'' methodology where a c yclic process of experiments with an ML tool, its data, data descripti on language and parameters attempts to optimize learning performance. Consultant-2 has been developed to support the use of Machine Learning Toolbox (MLT), an integrated architecture of 10 ML tools, and has evo lved from a series of earlier systems. Consultant-0 and Consultant-1 h ad knowledge only about how to choose an ML algorithm based on the nat ure of the domain data. Consultant-2 is the most sophisticated. It, ad ditionally, has knowledge about how ML experts and domain experts pre- process domain data before a run with the ML algorithm, and how they f urther manipulate the data and reset parameters after a run of the sel ected ML algorithm, to achieve a more acceptable result. How these sev eral KBs were acquired and encoded is described. In fact, this knowled ge has been acquired by interacting both with the ML algorithm develop ers and with domain experts who had been using the MLT toolbox on real -world tasks. A major aim of the MLT project was to enable a domain ex pert to use the toolbox directly; i.e. without necessarily having to i nvolve either a ML specialist or a knowledge engineer. Consultant's pr incipal goal was to provide specific advice to ease this process. (C) 1995 Academic Press Limited