Financial Early-warning System Of Listed Company Based On Biased Support Vector Machine

The enterprises will face with various risks and cope with the financial crisis during thesurvival and development. Especially the outbreak of the financial crisis in2008and the currentEuropean and American economic growth is weak, resulting in numerous cases of China’sEnterprise Bankruptcy. Therefore, in the current economic context, the financial position of theenterprise warning is particularly important. The current enterprises have become an urgent need tobuild a financial crisis early warning system.At present，the research on financial crisis early-warning in China is in a infancy．A lot ofempirical researches draw on traditional methods abroad. In recent years, machine learning is animportant part of the research in the field of artificial intellige nce and pattern recognition. Based onstatistical learning theory developed new machine learning method---support vector machine hasbecome a research hotspot of machine learning community, and made successful use in many areas,its processing of the small sample size, non-linear, high-dimensional data has shown characteristicssuperior to other methods. Due to corporate financial data in high-dimensional, nonlinearcharacteristics severely affect the accuracy of the early warning system, so this paper will applysupport vector machine algorithm to the enterprise financial crisis early warning model, the use ofthe excellent qualities in dealing with a small sample, non-linear, high-dimensional data to improvemodel accuracy and provide a valuable reference method for financial early warning of theenterprises.This article from domestic enterprises financial crisis proceed, first introduced the backgroundand significance of the topic and a review of the research status home and abroad; secondelaborated the theory of enterprise financial crisis and financial crisis warning, determined thesample data and warning variables of the model of financial crisis early warning, respectivelynormalized and principal component analysis of early warning variables corresponding pretreatment;and then used the standard C-SVM to build a model of enterprise financial early warning, forunbalanced sample data, predictions of the standard C-SVM can have some tendentious defects, sohave a proposal of improved support vector machine algorithm from the algorithm level and datalevel, we can use Biased-SVM algorithm to improve the standard C-SVM model by settingdifferent parameters of the punishment, we can use SMOTE method to generate a new sample forimbalanced data sets, so as to balance the different categories of data, and thereby improve theaccuracy of the classifier; at last we have a train of various early warning model and inspect theeffect of prediction. The empirical results show that: the training effect of the SMOTE-Biased-SVM model offinancial early warning improved in this paper is ideal, and the accuracy rate of financial earlywarning of listed companies can reach84%.