Mining interesting regions using an evolutionary algorithm

Abstract

In this paper, we offer a new method to induce interesting knowledge from the relevant sets of data in databases for supervised learning. Thus, in this work, ELLIPSES is presented as a new method oriented to discover knowledge according to the expert's needs, by the detection of the most significant regions. The method essence is found in an evolutionary algorithm that finds these regions one after another. The expert decides which regions are significant and determines the stop criterion. The extracted knowledge is offered through two types of rules: Quantitative and Qualitative. The tool also offers a visualization of each rule by parallel coordinate systems. The ELLIPSES results are compared with C4.5 on UCI Repository datasets.