Advanced oblique rule generating based on PCA

Abstract

In the paper a new algorithm of oblique decision rule induction is presented. It starts from dividing classes into subclasses which is a clustering problem. Then, around each subclass the hyperrectangle is built, which edges are parallel to PCA determined directions. Each hyperrectangle represents single decision rule which conditions are hyperplanes containing hyperrectangle sides. In order to simplify the obtained model less important conditions are removed from the rule and then less important variables are also eliminated from the hyperplane equations. The algorithm was applied for real and artificial datasets. © 2014 Springer International Publishing.