Comparative Analysis Of Penalized Variable Selection And Its Application In The Credit Risk Of Credit Card

The development of Information Age brings a massive of data. All fields will face the challenge of big data, such as bio-medicine, machine manufacturing, economics, finance, machine learning and IT. Thus, it is a necessity to select the useful information from huge data, where variable selection is an important component for information selection and be-comes a hotspot of statistical research in recent dozens of years. In the field of economics and finance, with the increasing of credit card business in commercial banks, current credit scor-ing mechanism cannot judge its risk effectively and efficiently because of nonsymmetrical credit information. However, adding too much user information will cause biased estimation because too much variables, makes more complex computations and a less stable credit risk of credit card model. So, it is significant to apply variable selection method into credit risk of credit card assessment. However, there is few research on penalized variable selection. Thus, this paper will discuss the comparative study of penalized variable selection, and apply it to credit risk of credit card assessment.First of all, the paper will classify penalized variable selection, summarize the charac-teristics, difference, advantages and disadvantages of several of variable selection methods. Then, we use different variable selection methods to do Monte Carlo simulation under dif-ferent data structures, coefficients and sample sizes, where data structures include with and without grouped conditions and variable correlation, coefficients include signs and zero ar-gument in the group, sample sizes include n<p and n> p conditions. The results of simulation compares and analyze penalized variable selection based on different variable se-lection and coefficient estimation conditions though FNR, FPR and model error, and comes up with suggestions of penalized variable selection methods on different conditions. Sim-ulation shows, Group Bridge in bi-level method has a good effect of variable selection and coefficient estimation under different data types.In addition, the paper will apply main penalized variable selection methods into credit risk of credit card assessment, comparison and find out that the result of penalized variable selection methods is better than normal Logistic model and stepwise regression model in both training test and test set. And we find out that13aspects of information, including installment, overdue, gender, working years, education, career and emergency contact, have an important impact on credit risk of credit card assessment.