تشخیص کلاهبرداری صورت های مالی و انتخاب ویژگی با استفاده از تکنیک های داده کاوی

Recently, high profile cases of financial statement fraud have been dominating the news. This paper uses data mining techniques such as Multilayer Feed Forward Neural Network (MLFF), Support Vector Machines (SVM), Genetic Programming (GP), Group Method of Data Handling (GMDH), Logistic Regression (LR), and Probabilistic Neural Network (PNN) to identify companies that resort to financial statement fraud. Each of these techniques is tested on a dataset involving 202 Chinese companies and compared with and without feature selection. PNN outperformed all the techniques without feature selection, and GP and PNN outperformed others with feature selection and with marginally equal accuracies.

Financial fraud is a serious problem worldwide and more so in fast growing countries like China. Traditionally, auditors are responsible for detecting financial statement fraud. With the appearance of an increasing number of companies that resort to these unfair practices, auditors have become overburdened with the task of detection of fraud. Hence, various techniques of data mining are being used to lessen the workload of the auditors. Enron and Worldcom are the two major scandals involving corporate accounting fraud, which arose from the disclosure of misdeeds conducted by trusted executives of large public corporations. Enron Corporation [17] was an American energy company based in Houston, Texas. Before its bankruptcy in late 2001, Enron was one of the world's leading electricity, natural gas, pulp and paper, and communications companies, with revenues amounting to nearly $101 billion in 2000. Long Distance Discount Services, Inc. (LDDS) began its operations in Hattiesburg, Mississippi in 1983. The company's name was changed to LDDS WorldCom [18] in 1995, and later it became WorldCom. On July 21, 2002, WorldCom filed for Chapter 11 bankruptcy protection in the largest such filing in US history at that time. Financial statements are a company's basic documents to reflect its financial status [3]. A careful reading of the financial statements can indicate whether the company is running smoothly or is in crisis. If the company is in crisis, financial statements can indicate if the most critical thing faced by the company is cash or profit or something else. All the listed companies are required to publish their financial statements every year and every quarter. The stockholders can form a good idea about the companies’ financial future through the financial statements, and can decide whether the companies’ stocks are worth investing. The bank also needs the companies’ financial statements in order to decide whether to grant loans to them. In a nutshell, the financial statements are the mirrors of the companies’ financial status. Financial statements are records of financial flows of a business. Generally, they include balance sheets, income statements, cash flow statements, statements of retained earnings, and some other statements. A detailed description of the items listed in the various financial statements is given below:

This paper presents the application of intelligent techniques to predict financial statement fraud in companies. The dataset consisting of 202 Chinese companies is analyzed using the stand-alone techniques like MLFF, SVM, GMDH, GP, LR, and PNN. Then, t-statistic is used for feature subset selection and top 18 features are selected in the first case and top 10 features are selected in the second case. With the reduced feature subset the classifiers MLFF, SVM, GMDH, GP, LR, and PNN are invoked again. Results based on AUC indicated that the PNN was the top performer followed by GP which yielded marginally less accuracies in most of the cases. Also, the results obtained in this study are better than those obtained in an earlier study on the same dataset. Ten-fold cross-validation is performed throughout the study. Prediction of financial fraud is extremely important as it can save huge amounts of money from being embezzled. Our study is an important step in that direction that highlights the use of data mining for solving this serious problem. With regards to the future research directions, we can extend this work by extracting ‘if–then’ rules from different classifiers. These rules can be helpful for easy understanding of the prediction process for the end user because they make the knowledge learnt by these techniques transparent. This type of knowledge elicitation can help in providing early warning. In addition to the data mining techniques used in this research, hybrid data mining techniques that combine two or more classifiers can be used on the same dataset. Also, text mining algorithms for sentiment analysis of the textual description of the financial statements can be used together with data mining algorithms for assessing the financial items in the financial statements to provide better prediction of financial statement fraud.