The Research On Key Techniques Of Financial Fraud Identification Based On Dynamic Combination Of Classifiers

In the modern market economy, the financial frauds of companies emerge in endlessly, that produces a tremendous impact on stock market and triggers an unprecedented credit crisis. Therefore, it is especially important to identify the financial fraud phenomenon for the companies. And there are high academic value and broad application foreground for applying classification techniques in data mining to analyze, compute and process the financial data, thus potential valuable information and rules could be mined which can help investors and accountants to easily confront various complex behaviors on the financial data. Currently, the researches on classification techniques applied to financial fraud identification have just developed. Moreover, there are quite a few problems that applying existing classification methods directly to financial fraud identification. Consequently, exploring effective classification methods suited to financial fraud identification has important real sense.The target, significance and status of the research are introduced in this paper. And an approach for dynamic combination of classifiers based on clustering division based on clustering division is proposed according to the characteristic of the financial fraud data and the problems in the existing ensemble methods. In order to evaluate the classification performance, we employ real financial data set of companies in the experiment. Meanwhile, a financial fraud identification system is designed and implemented using OOD technique.The main work of this paper is stated as follows:1. Abroad and domestic research statuses of dynamic combination of classifiers as well as fraud identification are reviewed. In addition, basic steps of financial fraud identification based on classification techniques are illustrated in details. Furthermore, some frequently-used classification techniques are talked about.2. A decision tree construction method based on rough set theory called S_D_Tree is put forward, which employs attribute significance instead of information gain ratio as the selecting criteria of testing attributes. Meanwhile, Failnode-prune pruning strategy is introduced into the construction process in order to get a reduced decision tree.3. An approach for dynamic combination of classifiers based on clustering division called DCC-CD is proposed. Firstly, PAM clustering algorithm is used to divide and reorganize the training samples for resolving the problem brought by unbalanced class distribution in financial data. Then, base classifiers are trained using the reorganized data sets and S_D_Tree learning scheme. Finally, predictions from base classifiers are dynamically combined to get the final result.4. DCC-CD is applied in the financial fraud identification. A feature selection method based on GA is raised to obtain the optimal feature set. And the classification performance of DCC-CD in practical application is evaluated. Meanwhile, a financial fraud identification system is designed and implemented using OOD technique.