Applying Bayesian Network, Decision Tree, Rough Set Theory, and Back Propagation Network Analysis for Detecting Accrual Earnings Management

• Main Authors: Wang, Yi-Cheng, 汪奕丞
• Other Authors: Chi, Der-Jang
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/td74j6

碩士 === 中國文化大學 === 會計學系 === 102 === Enterprise reliability in financial statements and management manipulated ac-cruals formed earnings management behavior has been a major topic of accounting. Previous studies in accrual earnings management are using traditional regression model. In recent years, a number of scholars have been doing research using data mining methods for accruals earnings management and accuracy was improved, but it is not enough to complete the whole literature. Therefore, this study use data mining method in the decision tree and back-propagation neural network to make predictions, hoping to achieve a more accurate detection mode. This study attempts to use CHAID decision trees, rough sets and Bayesian Net-work in the first stage of the variable filter. Further use of back-propagation neural network and C5.0 decision tree to modeling detect whether a company has a serious manipulation of earnings. The empirical results show that rough sets theory screening method with back-propagation neural network has the best performing, and it’s accuracy rate is 96.82%.