Elastic Multi-stage Decision Rules for Infrequent Class

Abstract

Typically, decision trees are used to represent knowledge by rule generation. To have a better understanding of the rules, it is sometimes necessary to minimize the number of nodes by minimizing the depth of the tree. This study optimizes the depth of the tree by minimizing the number of nodes. Rules that are generated using either decision trees or class association mining are from the major class of the dataset. To enable rules to be created for the infrequent class, this study uses an elastic method, Elastic Multi-Stage Decision Methodology (EMSDM), to create rules for the infrequent group. EMSDM is elastic in that it expands and contracts to accommodate the characteristics of the dataset. In addition, the data analysis occurs in stages: clustering, minimizing the depth of the decision tree, and association mining, to increase the ability of EMSDM to find infrequent class rules. EMSDM shows promise to find infrequent class rules with increased accuracy.