Mining multidimensional data through element oriented analysis

Abstract

Mining multidimensional data has two major concerns. One is how to select the most salient attributes and another one is how to guarantee the precision of mining results. This paper introduces a novel approach to mine multidimensional data through Element Oriented Analysis (EOA). In our approach, each observational data is considered to be comprised by two essential elements, the structure elements and the numerical elements. EOA firstly targets Structural Element Pattern (SEP) that is an aggregation of the structural elements. The successful SEP will be referenced by a Numerical Element Pattern (NEP) that is composed of the numerical elements. Given the results from both SEP and NEP, a global discriminant will be created for the efficient evaluation by the consequent data. In this paper, publicly available Turkish bank records are analyzed in an experiment that demonstrates the practical utility of our approach. © 2008 Springer Berlin Heidelberg.