Study On The Personal Credit Evaluation Based On The Modified Na(?)ve Bayesian Method

With the rapid development of the economy in China, the living standards of citizens have been improved gradually. Personal consumption and investment philosophy are changing along with the better-life styles. As a result, personal credit consumption is expanding unceasingly and gradually becoming the fundamental way of personal consumption. In recent years, the rise of Internet financial greatly promotes rapid growth of personal consumption credit. But, nowadays, the further growth of personal consumption credit is greatly restricted by the backward credit system in our country. As one of the most important part of credit system, personal credit assessment system becomes a hot spot in social and academic concern and research.Many scientific and reasonable personal credit evaluation model are proposed by domestic and foreign scholars, including Discriminate Analysis, Logistic Regression, K-Nearest Neighbor Method, Naive Bayesian, Classification Tree, Decision Tree, Artificial Neural Network, Genetic Algorithm and Support Vector Machine(SVM), etc. The Naive Bayesian method, one of the most important classification algorithm in machine learning field among them, has been employed most because of its simple structure, high classification accuracy, efficiency and systematic mathematics theory foundation. It is truly one of the best classifiers at present. But in practice, even the independent-condition assumption for a simple Bayesian is nearly impossible to guarantee, of which the classification performance will be inevitably worse.In this paper, Naive Bayesian classification's performance is improved by relaxing the condition-independent assumption firstly. In addition, three personal credit evaluation models are formulated based on Bayesian method of principal component analysis?attribute-weights and rough-set, respectively. Then, personal credit evaluation index system is built based on both selecting principle of personal credit evaluation index system and current personal credit assessment index system.Finally, German personal credit evaluation data in UCI machine learning public data set is selected as our data source. Through empirical examples, the effectiveness of the proposed three models is proved. And their improvement on naive Bayesian method is confirmed.