Detecting financial statement fraud: A comparative study using data mining methods

作者：Song Xinping^*; Ge Yan

来源：International Review on Computers and Software, 2012, 7(4): 1778-1783.

• Back-propagation neural networks
• Chinese companies
• Classification accuracy
• Comparative studies
• Data mining methods
• Data sets
• Error rate
• Financial statements
• Hybrid classifier
• Individual models
• Prediction accuracy

In this paper, we propose data mining methods for detecting financial statement frauds. We conducted an experiment on a data set from Chinese companies. Four individual classifiers-Logistic Regression (LR), Back-propagation neural network (BPNN), Decision tree (DT), Support vector machine (SVM) and a hybrid classifier are tested and compared. As the result indicates, the hybrid classifier outperforms all the individual models used in terms of prediction accuracy and composite error rate, whereas SVM outperforms LR, BPNN, and C5.0 DT in term of classification accuracy and good generalization.

• 出版日期2012