The Analysis Of Credit Classification Model Based On The Data Mining

With the rapid development of financial industry,credit classification model becomes particularly important.The purpose of credit classification model is distinguishing good customers and bad customers through the basic information which the financial collect,such as age,income,marital status,etc.A good credit classification model can be a forceful support in financial business decisions and reduce the risk in commercial communication.Developing and implementing a credit classification model can avoid time and resource consuming.This article establishes a scientific and effective credit classification model based on UCI database.Eventually,this model can improve the efficiency of financial industry and avoid capital wasting.The development of data mining technology offers the possibility of quantitative analysis to mass data.Data mining technology also becomes the indispensable financial instrument.In this article,we introduce spectral clustering algorithm.Clustering algorithm is a common data mining technology which is suitable for data classification.Traditional clustering algorithm can only deal with low-dimensional datasets,not sensitive to high-dimensional datasets such as credit datasets.Distant calculating is very important in traditional spectral clustering algorithm,this paper uses the orthogonal distance replacing Euclidean distance when using spectral clustering algorithm according to these above situations.Then we compare the improved spectral clustering algorithm with the traditional spectral clustering algorithm and the traditional clustering algorithm such as the k-means clustering algorithm.The results have turned out that the improved spectral clustering algorithm has better accuracy.The spectral clustering algorithm based on the orthogonal distance has increased the accuracy by 15 percent average than K-means clustering algorithm.The spectral clustering algorithm based on the orthogonal distance has increased the accuracy by 34 percent average than traditional spectral clustering algorithm.