Filtering Decision Rules with Continuous Attributes Governed by Discretisation

摘要

The paper presents research on selection of decision rules with continuous condition attributes while exploiting characteristics of these attributes obtained by supervised discretisation. The considered features were split into categories corresponding to numbers of intervals required for partitioning of their values, and this information was next used to divide the sets of rules by their conditions falling into specific categories. Also to each group of variables there was assigned some weight, basing on which several rule quality measures were calculated. They enabled filtering rules meeting requirements with respect to performance.