Credit Scoring Using Sparse Bayesian Learning

With the rapid development of China’s economy, people’s consumption and investment sense are constantly changing, and personal credit transaction under the market economy has become the most important form of personal transaction. According to the consumer credit data from the People’s Bank of China in2013, the total scale of credit transaction has been expanded, and has maintained annual growth rate of24%since2010. At the same time, the rapid development of consumer credit has also increased the credit risk of commercial banks. On the one hand, commercial banks are striving to develop personal credit business, on the other hand they are also trying to avoid risk and taking into account the income and security. In the global financial crisis of2007, ignoring and underestimating the credit risk of the individual is an important incentive. As the crisis gradually subsides and the global economy recovers slowly, the credit institutions, regulatory authorities and research scholars pay more and more attention to the individual credit risk. Therefore, a reasonable and effective personal credit evaluation can regulate and promote the development of commercial banking business, on the other hand personal credit evaluation is also conducive to the normal operation of the commercial banks and the economic security of the country.Credit scoring plays an important role in risk managements and it is a typical classification task in pattern recognition. Domestic and overseas scholars do a lot of research from the perspective of statistics, artificial intelligence and machine learning. Traditional models suffer from low accuracy of classification and poor interpretability of selected features. To solve this problem, a novel model using Sparse Bayesian Learning(SBL) to evaluate personal credit risk (SBLCredit) was proposed in this thesis, which utilizes advantages of SBL to get as sparse as possible solutions under the priori knowledge on the weight of features and leads to both good classification performance and effective feature selection. The main contents of this thesis are as fllows.The first chapter is introduction. The second chapter is research status and research methods of individual credit assessments. The third chapter is the theoretical basis of SBLcredit. The fourth chapter is personal credit evaluation system. The fifth chapter is data preprocessing and SBL-credit model building. The sixth chapter is empirical results comparison. The last chapter is the summary and outlook.The main contributions and innovations of this thesis are as follows:(1) Reviewing the status and methods of individual credit assessment up to now.(2) Introducing SBL to individual credit assessment areas and proposing a personal credit evaluation model-SBLCredit,which is the largest contribution and innovation of this paper.(3) Extensive experiments on Germany, Australia and UK real credit data sets proved the feasibility and effectiveness of SBLCredit from many aspects and views.