Scholarly record

CLUSTERING METHOD BASED ON ITERATIVE NODES PAIRING IN A DELAUNAY TRIANGULATION NET AND ITS APPLICATION FOR BUSINESS DATA ANALYSIS

Janusz Morajda, P. Woloszyn

Abstract

Clustering techniques are frequently applied in business and economy, for example in customer classification, enterprise grouping and in other economic data analysis. This paper presents a new clustering method, based on Delaunay triangulation and iterative aggregation of points distributed in the feature space. In the first part of the article we propose a clustering algorithm that utilises this concept. The algorithm iteratively connects specially selected pairs of nodes in the triangulation mesh and leads to systematic reduction of the mesh, together with creation and development of clusters. The procedure can be then classified as a hierarchical, agglomerative clustering method. Particular stages of the algorithm have been described and discussed, and its pseudo-code has been presented. The algorithm has been implemented in form of computer program written in C++. So far, the program can be applied to 2- or 3-dimensional problems (i.e. to datasets with only 2 or 3 numerical features), however in case of n variables (n > 2) the principal component analysis (PCA) tool used as a data preprocessing method can provide dimensionality reduction and decorrelation of variables. The method (together with PCA) has been applied to a selected business problem, concerning clustering of a set of enterprises and then identification of bankruptcy risk classes of those companies. We have obtained promising results that have been visualised and discussed. Directions of further research include generalising the algorithm to any number of variables, based on n-dimensional Delaunay tessellation.

Publication Impact Profile

No metrics available.

PLUMX see details

DOI resolver: doi.org/10.5593/sgemsocial2018/1.5/S05.017

Publication details

Back to publication list