Abstract:
Investors, shareholders, financial analysts and other stakeholders pay special attention to the figure of profit reported in financial statements as one of the main factors in decision-making. Hence the recognition of earnings management in the financial statements is very important.  The main objective of this research is to investigate the effectiveness a model to specify the earnings management level by combining two data-mining techniques including a Bayesian networks and a decision tree. To do so, 24 variables have been used as the independent variables affecting earnings management and discretionary accruals as the dependent variables replacing earnings management.In this study the recommended method by Dechow et al (2012) was used to measure the earnings management. The research's statistical population included listed companies on the Tehran stock exchange, active in automotive and parts manufacturing industry and device constructions during years 88-93. The research findings indicate that the hybrid model of Bayesian networks with the C5.0 decision tree has a high capability to characterize the level of the earnings management in the automotive and parts manufacturing industry. The most important known variable by the hybrid model to identify the level of the earnings management in the automotive and parts manufacturing industry is performance threshold variable with the significance ratio of 100%.