The Facticity Checking Model Of Financial Statements Based On Support Vector Machine

The financial statements are written documents,which reflecte a comprehensive accounting entity in a certain period of time and the Mainland financial position, operating results and financial changes, and which are an important basis for making decisions to investors and creditors.In early 21st century, listed companies seemed have become the global financial fraud the focal point, even be considered to be an important feature of this era. In 2001, following the Enron Corporation (Enron), a respect for the years of the energy trading industry giants financial frauds in the capital market, a heavy bomb dropped, then pull the United States for the past few years, the biggest financial fraud scandals : Arthur Andersen, global communications, Xerox and other major companies have world-renowned escape. United States,which had always been regarded as a model for the capital market and corporate governance structure , has began to been serious questioned . The deterioration of Investors'continue confidence crisis had seriously threatened the United States economy.Since the early 1990s in the 20th century when meanwhile the founding of the Shanghai and Shenzhen Stock Exchanges ,China had made rapid development of the securities market from scratch. Meanwhile, the accounting information disclosure system, as the stock market to achieve "open, fair and just" principles of protection With the continuous improvement of accounting systems, securities regulators and the constant strengthening of the development of a certified public accountant. gradually established and developed. However, when we affirm our achievements, we should also clearly see that China's disclosure of accounting information should not be blindly optimistic about the current situation. The corrupt financial reports of listed companies in China's stock market has never been fractured.The corrupt financial statements caused significant losses to the companies. The companies were forced to declare bankruptcy or dropped in seriously corporate image in the ruined capital and product markets and financial difficulties. Meanwhile, the fraud to make financial statements more corrupt stakeholders loss, holding the hands of shareholders in the share price plummeted and the banks would face a huge bad debt losses, accounting firm to pay hundreds of millions of dollars and may even collapse. Even more serious is that the fraud may lead the public to the financial statements of listed companies, intermediaries, even undermine confidence in the capital market, lost, thus affecting the development of the entire economy.If we can construct a classification model, which on the basis of the financial statements of listed companies regularly publish figures on the financial operations can be classified veracity of the report itself assessment is very meaningful. The model can be used to control the selection and implementation of a key survey of high-risk companies; This model can be registered accountants auditing contract and the signing of the audit planning stage to assess the risk of fraud customers; Retail investors can also use the model to a high probability of fraud screening company in order to avoid possible investment losses.Researchers try to use machine learning ( pattern recognition ) identification algorithm to construct a model of financial fraud. The common used algorithms is K-nearest neighbor, decision trees, neural networks. These models have made some experimental results, these algorithms are based on the probability measure itself and the law of large numbers. At the moment we know little samples of fraud, and the indicators reflect the financial position, is doubt about the reliability of the algorithm. In particular neural network algorithm, the thousands of samples is not difficult to train a good model.Researchers have introduced such a support vector machine algorithm for pattern recognition applied to the small sample size, However, due to limited selection of targets and the quadratic linear programming SMO into convex programming algorithm. SMO support vector machine is based on linear programming algorithm, the algorithm will be the second convex programming into linear programming, the sample size is large, high-dimensional algorithms. In the small sample size, low-dimensional data from the training, they do not yield good results.My research start from the theory of financial fraud, fraud-depth analysis of the motives and financial characteristics of the environment. Aggregate financial position also reflects the close relationship with the 37 financial indicators of fraud.I choose the support vector machine algorithm of the kernel-based . This ingenious methods can solve these two problems: The use of nuclear function theorem make us do not need to know the explicit expression of non-linear mapping; As in the study of high-dimensional feature space planes using the linear method, the linear model with almost no increase in the complexity of calculation, in part to avoid the "dimension of the disaster." This algorithm can be regarded as an excellent machine learning and pattern recognition algorithms for small size and high-dimensional data.I use the kernel-based support vector machine algorithm to selected for a three-dimensional evaluation of the entire portfolio of training of the 37 indicators. Then select the best indicator of several groups, do the entire portfolio evaluation indicators for the entire peace keeping training. Finally I get a stable and satisfactory evaluation model.This research and results has positive significance for the further research of the construction of financial fraud recognition model.