How to protect investors? A GA-based DWD approach for financial statement fraud detection
Xinyang Li, Wei Xiang Xu, Xuesong Tian · 2014
As one type of the financial fraud, financial statement fraud has not only led to a huge loss for individual investors and financial institutions, but also impacted the overall stability of the whole industry. This paper used financial and textual features extracted from annually submitted 10-k filings and combined data and text mining techniques for detection of financial statement fraud. When the dimension of samples is larger than the sample size, namely high dimension low sample size (HDLSS), distance weighted discrimination (DWD) model, which has a good generalization performance in HDLSS contexts, is used to detect financial statement fraud. We also adopted genetic algorithm to improve the performance of classifiers, including DWD, Support Vector Machine, Back Propagation Neural Networks and Decision Tree for feature selection and parameter optimization. Compared with other GA-based classification models, the proposed GA-based DWD model achieved relatively high classification accuracy with fewer input features, which proves that this model is a promising tool for detection of fraudulent financial statements.