The detection of misstated financial reports using XBRL mining and intelligible MLP

Abstract

Considerable effort has been devoted to the development of integrated software to assist in the detection of financial misstatements. Despite this, the use of such tools has been sparse due to the opacity of the resulting output and the complicated task of importing the financial data they require. This article presents a conceptual framework for modelling financial statements that leads to significantly improved performance, allowing a Multi-layer Perceptron with a modified learning method to form internal representations that can be easily interpreted by financial analysts. The article dis-cusses the use of XBRL data extraction from the web, showing how a judicious selection of accounts can help solving the cumbersome problem of im-porting data. The resulting tool makes the detection of financial misstatements both understandable and easy.