The Self-Organizing Map in Selecting Companies for Tax Audit

Abstract

Today, Tax Authorities receive the tax reports from companies to a large extent in digital form from the companies in Finland. Most of the tax reports are processed routinely i.e., a computer program checks that the taxes paid in advance are the correct ones and if not, the company either receives a tax return or is asked to pay the difference and there is no need for a tax audit. However, there is a small percentage of companies that need it. Most of these companies – for some reason – have not reported all their income items or have reported cost items that do not belong to their report. This could be unintended or it could be fraud. The problem is to find this percentage from the mass of tax reports. So far, the tax auditors or tax inspectors have used their past experience and posed queries to the data base, where the reports are stored, to find the ones that need a tax audit. This is not necessarily the most effective way of finding the tax reports that need a tax audit. Different data mining tools might aid in this process and make the selections of companies that need tax audit more effective. The aim of this paper is to investigate how well an unsupervised neural network method – the self-organizing map (SOM) – can perform in the task of finding the companies that need to be tax audited. SOM is a data driven approach without a need to have predefined rules or sets of values. A real data set is used and the results are compared to the results that the tax inspectors have received with their methods.KeywordsFeature PlaneInspection ResultTaxation DataInput Data VectorIncome ItemThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.