Visualization of Patterns in Accounting Data with Self-organizing Maps

Abstract

Neural networks are data driven methods. They provide additional information to the decision process as might be left hidden otherwise. Neural networks have already been applied in many different business areas; and they can be used for prediction, classifying, and clustering. They can learn, remember, and compare complex patterns. This chapter shows how a neural network, especially Kohonen’s self-organizing map (SOM), can be used in visualization of complex accounting data. The SOM is used for clustering ten years of monthly income statements of a manufacturing firm. The purpose is to show how the data sets of various accounts and years form their own groups. We found that the SOM can be a visual aid for classifying and clustering data sets, and that it reveals if some cluster contains data that a priori should not be in it. Hence, it can be used for signaling unexpected fluctuations in data. Furthermore, the SOM is a possible technique embedded in the continuous monitoring and controlling tool.