Principal Components and the Accuracy of Machine Learning
Induction algorithms could be positioned somewhere between witchcraft and science since extracting rules (the essence of data semantics) from an incomplete set of examples is a devil’s task. On the other hand there always has been some romanticism in the interpretation of eigenvalues. Authors of this paper try to show that the abovfe mentioned notions work together quite well. Quite surprising results have been achieved. Not only the number of decision variables can be reduced but, in most tested cases, the prediction accuracy improves (despite the reduced number of variables). The results could be implemented as a preprocessor for most of the induction algorithms employed for an automatic rule generation (one of the most important components of any modern expert system).
KeywordsExpert System Induction Algorithm Decision Tree Induction Multivariate Discriminant Analysis Knowledge Representation System
Unable to display preview. Download preview PDF.