Support Vector Machines And Its Corporate Credit Risk Rating Application

Support Vector Machine (SVM) is based on statistical learning theory developed from the new machine learning theory. Compared with the traditional machine learn-ing methods, support vector machine method Handle relatively small sample size and sample data to adapt to deal with the advantages of the situation can not be separated, It VC dimension theory and structure of the wind Risk minimization principles, the model complexity and learning ability to achieve a good balance between. Make support vector machine was a very good generalization ability.Support Vector Machine learning in the key definition of consistency refers to the training set goes to infinity, the empirical risk converges to the optimal value of the optimal value of the real risk. Only when conditions meet only to ensure consistency of ERM principles of the best ways to come. When the sample infinity, it can be Wanted optimal results, the expected risk close to the minimum. Classification using support vector machines one of the major advantages is that it not only into consideration the Posterior risk, but also noted that the use of the generalization ability of classification, namely We can see from the above equation R(w) is composed of two parts, the first part of R, that is, empirical risk. The second part is called Confidence range of experience, he reflects the difference between risk and real risk of upper bound, its structures and functions VC dimension and the number of samples were related.One on the VC dimension [3] there is defined as an indicator function for the set, if there exists h samples Can be a function of concentration of all functions by all the possible 2h kinds of methods are divided into two categories, Claimed that this func-tion sets the number of h be able to sample a collection of scattered samples. Direct function sets the VC Dimension refers to the concentration of using this function, the function can broken up the largest number of samples of the sample set. If the number of samples for any one always be able to find a sample set by this function sets the Broken up, then the function is set on the VC dimension is defined as infinity.In the support vector machine theory is accurate for classification plays an im-portant role in nuclear function, commonly used kernel functions are the following:(1) Linear kernel function:(2) polynomial kernel function for any positive integer d, c> 0 there(3) Gauss RBF kernel function:(4) multi-layer perceptron kernelThe kernel functions in the application of support vector machines have played an important role.With more and more people support vector machine-depth study of various de-formation algorithm came into being, Such as:C-SVM, C-SVM algorithm is a second slack variable, C-SVM algorithm K-th slack variable, BSVM algorithm, Bv-SVM al-gorithm, v-SVM, W-SVM, RSVM, LS-SVM, PSVM algorithm Fuzzy support vector machine algorithm. These methods mainly through increasing function items, change the variable or factor to achieve.In recent years, support vector machine, not only in theory made great progress, but also in the same algorithm and applications made considerable progress. In the settlement, such as text recognition, spam filtering, Recognition, medical diagnosis, Image Processing, Natural language processing, and credit rating classification, sup-port vector machines have achieved good results. Especially in the corporate credit rating, the support vector machine has been more and more attention.Corporate credit rating is the bank lending the core issue, the rating results for banks to develop relevant policy issues and how to get more profit with significant meaning. Credit rating with this know that the safety, reliability and its robustness is directly related to the success of each loan business. The assessment of domestic bank credit is still largely remain in the traditional ratio analysis method stage, with the de-velopment of market economy simply Yikao traditional financial theory and corporate management to make judgments on the brain is obviously not enough. This triggered a search for new theories, new tools, using a more effective way for enterprises to analyze and judge the state of the economy needs. This article is based on this to ex-plore the use of support vector machines to corporate risk rating. The so-called credit rating means an index system based on scientific, using rigorous analytical methods, using simple text symbols Graded units that carry on financial responsibilities Ability and trust degree objective and impartial evaluation and to determine its credit rating of an economic activity.General corporate credit rating of the content is very extensive, mainly inte-grated, simplicity, timeliness and fairness and so on.This numerical experiment used SVM package LIBSVM program is the de- velopment of Taiwan University, Dr. Lin Zhiren general support to the Amount of machine pattern recognition and regression of the package. The financial indicators of the experiment by the listed company's balance sheet and profit distribution state-ment was arrived at. Our main financial indicators used in the profitability, financial ratios and liquidity. Capacity in the profit margin we have selected assets, operating profit margin. In the financial ratios selected are current liabilities ratio and total lia-bilities ratio. Choice of the liquidity of financial indicators are cash ratio, cash asset ratio, quick ratio and current ratio. Then use the software LIBSVM classification and prediction, and thus derive the kernel function in predicting outcomes.The data used in this chapter are taken from the Star of securities published on the website in September 2008 to September 2009 the four quarters of the financial statements and profit distribution statement. Including real estate classes, engineering and construction category, Computer classes and wine food, etc.30 categories of non-ST 11 ST companies and related companies.As the business strength and industry analysis in the evaluation of a Credit status of enterprises is very important, so I chose corporate strength of the company are Large private enterprises and state-owned enterprises, but as far as possible in the industry on a wide range. Because the original intention of this article and assets from the financial statements of public income statement data test the accuracy of SVM classification, we select the parameters in the time pay more attention to Objective report on financial data, and spent a few such as:leadership quality, the number of independent directors, and other indicators.In this experiment, the selected kernel functions are linear kernel, sigmoid nu-cleus and polynomial kernel. Accurate prediction of different kernel functions were: 77.1429%,71.4286%and 77.1429%. Linear prediction of nuclear and nuclear poly-nomials Accuracy rate under the prerequisite of a good polynomial kernel is 41 times the number of iterations, while the number of iterations of linear kernel it reached 110. The experimental results show polynomial kernel functions in the credit classification efficiency and accuracy of forecasts is the best.