Multistrategy Learning from Engineering Data by Integrating Inductive Generalization and Genetic Alg

J. Bala,P. Pachowicz,K. D. Jong

Abstract:The Sil.C amJ cOlOplexity of most engineering data and an inherent presence of n(lise in Ihem pose significant uifficulties for traditional concept learning ~ystcms. This chapter presents a multistrategy system thaI addresses these issues. The sys­ tem integrates twO forms of learning-induclive gcm:ralilation and genelic algo­ rithms-in a closcd-loop fashion in order to achieve rohust concept learning c;ap;a­ bililie); for a variety of difficult engineering-oriented c1assificalion lask~. TIle learning process cycles through two phase~. In the first phase. ini tial concept descriptiolls arc acquired hy running a !loise-tolerant !!xtensioll ()f the familiar AQ ruk inductioll system. The resulting concept descriptiolls arc not. however. optimal from a pcrformau~e point of view because of the ~trong AQ bias to generate sim­ ple. cognilivcly oriented descriptions. Therefore, in the seconu pha.'c:. the pcrtQr­