Research On Enterprise Credit Assessment Based On Kernel Principal Component Analysis And Support Vector Machine

Fundamentally speaking, market economy is kind of credit economy, whilecredit plays an essential role in the financial system to guarantee the normaloperation of market economy. A consummate credit system gives a greatguarantee of normal management of enterprises and rapid development ofnational economy. However, with the continuous development of marketeconomy, a range of issues triggered by the higher frequencies of credit-doubtin the credit crisis have become major obstacles constraining China’seconomic development. Thus, the need to establish an effective creditmanagement system to prompt China’s economic development has becomeextremely anxious, in which the key is to do a good job of enterprise creditassessment effectively.In recent years, scholars have gradually applied Neural Networks,Support Vector Machines and other smart models to the field of enterprisecredit assessment and gained encouraging results. However, previous studiesfocused more on the intelligence assessment model, while the selection ofcredit assessment index system has not been fully explored. This paper makesa further analysis on the basis of previous studies. It brings up a variety offactors to build up a credit assessment index system based on scientific andrational principles of selection of indicators. As the Kernel Principal ComponentAnalysis is kind of effective newly-developed extraction method to deal withnon-linear problem on the basis of Principal Component Analysis. Thus, in theprocess of modeling, this paper makes use of Kernel Principal ComponentAnalysis to extract the constructed index system, which can effectivelyovercome possible side effects that brought by excessive indicators or by thecorrelation of indicators that effect the performance of credit assessment.SVM is kind of new machine learning method based on StatisticalLearning Theory. By virtue of its superior performance and good generalizationability, SVM has been widely used in various fields and achieved good results.In essence, traditionary Support Vector Machine can only settle problems of two-class classification. However, in the practical application,multiple-class classification problems are more common, enterprise creditassessment in this paper as well. In this case, we should extend two-classSupport Vector Machines into multi-class classification Support VectorMachines with a combining strategic. As the Binary Tree multi-class SupportVector Machines can effectively overcome the shortcomings oftime-consuming and low accuracy in previous multi-class Support VectorMachine, this paper applies the new algorithm to accomplish enterprise creditassessment. Whether the parameter on each node of decision tree in SVM isfit or not directly determines the performance of the constructed model,consequently to the outcome of the assessment. As the Chaotic ParticleSwarm Optimization would avoid the problem of easily drop into partialminimum which occurs in Particle Swarm Optimization and effectively increaseaccuracy training and testing in Support Vector Machine. Its performance isobviously better than other algorithms, so this paper use Chaotic ParticleSwarm Optimization to automatically optimize the parameters, and then takeadvantage of the Binary Tree multi-class Support Vector Machine whichparameters have optimized and Kernel Principal Component Analysis appliedto enterprise credit assessment to achieve an effective assessment.