Detecting fraudulent financial statements for the sustainable development of the socio-economy in China: A multi-analytic approach

Abstract

Identifying financial statement fraud activities is very important for the sustainable development of a socio-economy, especially in China's emerging capital market. Although many scholars have paid attention to fraud detection in recent years, they have rarely focused on both financial and non-financial predictors by using a multi-analytic approach. The present study detected financial statement fraud activities based on 17 financial and 7 non-financial variables by using six data mining techniques including support vector machine (SVM), classification and regression tree (CART), back propagation neural network (BP-NN), logistic regression (LR), Bayes classifier (Bayes) and K-nearest neighbor (KNN). Specifically, the research period was from 2008 to 2017 and the sample is companies listed on the Shanghai stock exchange and Shenzhen stock exchange, with a total of 536 companies of which 134 companies were allegedly involved in fraud. The stepwise regression and principal component analysis (PCA) were also adopted for reducing variable dimensionality. The experimental results show that the SVM data mining technique has the highest accuracy across all conditions, and after using stepwise regression, 13 significant variables were screened and the classification accuracy of almost all data mining techniques was improved. However, the first 16 principal components transformed by PCA did not yield better classification results. Therefore, the combination of SVM and the stepwise regression dimensionality reduction method was found to be a good model for detecting fraudulent financial statements.