POST-PROCESSING OF CLASSIFIERS - KDD

Abstract

Data Mining potentializes the use of the large volumes of data stored by companies, discovering useful patterns to support decision making. However, it is common that the quantity of these discovered patterns makes the respective analysis unfeasible. This paper proposes and tests two strategies based on generalization to post-process discovered patterns in a decision tree format. Both strategies were tested on a decision tree discovered within a data range of 1,617 employees, having as class attributes the fact that they either were or weren’t awarded a sick leave (absence) for orthopedic treatment. The results achieved from the adoption of these two strategies show a reduction of around 60% of the patterns to be analyzed.

KDD; Post-processing; Decision tree; Generalization