Managing Complexity in Large Data Bases Using Self-Organizing

Managing Complexity in Large Data BasesUsing Self-Organizing Maps

Barbro BackTurku School of Economics and Business AdministrationP.O. Box 110, FIN-20521 Turku, Finlandemail:bback@abo.fiMikko Irjala

Turku School of Economics and Business AdministrationP.O. Box 110, FIN-20521 Turku, Finlandemail:mikko.irjala@utu.fiKaisa Sere*

Åbo Akademi University, Department of Computer Science,Lemminkäisenkatu 14, FIN-20520 Turku, FinlandKaisa.Sere@abo.fi*On leave of absence from University of KuopioHannu Vanharanta

University of JoensuuP.O. Box 111, FIN-80101 Joensuu, FinlandHannu.Vanharanta@abo.fi

Turku Centre for Computer ScienceTUCS Technical Report No 48September 1996ISBN 951-650-844-8ISSN 1239-1891AbstractThe amount of financial information in today's sophisticated large data bases is huge and makescomparisons between company performance - especially over time - difficult or at least verytime consuming. The aim of this paper is to investigate whether neural networks in the form ofself-organizing maps can be used to manage the complexity in large data bases. We structureand analyze accounting numbers in a large data base over several time periods. By using self-organizing maps, we overcome the problems associated with finding the appropriate underlyingdistribution and the functional form of the underlying data in the structuring task that is oftenencountered, for example, when using cluster analysis. The method chosen also offers a way ofvisualizing the results. The data base in this study consists of annual reports of more than 120world wide forest companies with data from a five year time period.

This paper is an extended version of our paper Data Mining Accambis Numbers Using Self-Organising Maps presented at Finnish Artificial Intelligence Conference in Vasa 20-23 August1996.

Keywords: Neural networks, self-organizing maps, data bases, benchmarking

TUCS Research GroupComputatinal Intelligence in Business1. IntroductionCompetitive benchmarking is an important company-internal process, in which thefunctions and performance of one company are compared with those of othercompanies. Financial competitive benchmarking uses financial information æ mostoften in the form of ratios æ to perform these comparisons. Financial competitivebenchmarking is utilized, among other things, as a communication tool in strategicmanagement, for example in situations where company management must gainapproval, from internal and external interest groups alike, for new functional objectivesfor the company.

Multivariate statistical methods have been used as a tool of analysis for companyperformance, bankruptcy predictions, stock market predictions etc., although mostly inresearch contexts. However, many problems have been reported concerning thesemethods. The two most important problems are the assumption on normality in theunderlying distributions and difficulties in finding an appropriate functional form for thedistributions. Moreover, results of analyses are difficult to visualize when there areseveral explanatory variables [Vermeulen et al., 1994].

Many researchers have addressed these problems: Trigeueros [1995] reports on severalstudies that have shown the existence of positive or negative skewedness in the ratiosand on different remedies to overcome these difficulties. He also explains the existenceof symmetrical and negatively skewed ratios and offers guidelines for achieving higherprecision when using ratios in statistical context.

Fernandez-Castro and Smith [1994] used a non-parametric model of corporateperformance to overcome the need for specification of statistical distribution orfunctional form. Vermeulen et al. [1994] presented a way to visualize the results withinterfirm comparison when the explanatory variable was explained by more than onefirm characteristic.

Vanharanta [1995] has used modern computer technology and built a hyperknowledge-based system for financial benchmarking. The system contains a data base withfinancial data on more than 160 pulp and paper companies worldwide. The amount offinancial information in this system is, however, so large that it makes comparisonsbetween companies difficult æ or at least very time consuming.

Artificial neural networks are a promising new paradigm in information processing.Originally, they were developed as computer analogues for the human brain [Hecht-Nielsen, 1991]. Artificial neural networks are able to learn the pattern of a system froma given set of examples; a feature which makes them very attractive. They areapplicable to such processes as classification, prediction, control, and inference[Rumelhart et al., 1986].