The aim of this study is to investigate the impact of various pre-processing models on the forecast capability of artificial neural network (ANN) when auditing financial accounts. Hence, the focus of this paper is on the pre-processing of the data. ANNs are selected for auditing purposes because they are capable of learning complex, non-linear underlying relationships. Therefore, they are used to model the dynamics and the relationships between account values in order to find unexpected fluctuations. This study uses a multi-layered neural network with the backpropagation algorithm. The artificial neural network model used in this study was built by using the financial statements of 31 manufacturing companies over four years. The values of the accounts were regarded as a timeseries. The data were pre-processed in four different ways. Firstly, all the data were scaled linearly. Secondly, the data were pre-processed linearly on a yearly basis. Thirdly, the data were pre-processed linearly on a company basis. And fourthly, the data were pre-processed on a yearly and company basis. The best results were achieved when all the data were scaled either linearly or linearly on a yearly basis.
Koskivaara, Eija, "Different Pre-Processing Models for Financial Accounts when Using Neural Networks for Auditing" (2000). ECIS 2000 Proceedings. 3.