Michal Draminski

Elements of constructive induction in INELENStar system

930

Abstract

Data mining is the search for relationships and global patterns that exist in large databases. This paper provides overview of database mining as the confluence of machine learning techniques and performance emphasis of constructive induction. Constructive induction reduces the sensitivity of an inductive algorithm to its vocabulary by enabling the algorithm to construct new variables. Constructive induction is a much faster way to search the space of possible vocabularies. This work present three algorithms implemented in INLENStar application. One of these algorithms is new "Apriori*" designed especially for INLENStar, and two well known algorithms "COBWEB" and "CLARA". Generally the three algorithms search for new variables using three difficult methods of clustering. The paper considers a large number of experiments, to study properties of the algorithms.

Keywords : data mining, machine learning, constructive induction.