The Research On Personal Credit Scoring Based On CBR Cycle Optimization

With the accelerating world economic integration process, risk control is moreimportant to commercial bank. As one of the essential parts of bank assets business,the level of personal credit is related to the general requirements of economycapital.Therefore, effective recognition and measurement for personal risk willbecome the key factor of reasonable risk control for the commercial bank.At present, most credit scoring models globally are built on statistics orartificial intelligence, which are all double-edged. Statistical models offer tests butfall short of accuracy, while artificial intelligence achieves high accuracy but fallshort of explanatory function. Furthermore, these models are challenged by biasedsample and dynamic changing of credit samples which need to be solved urgently.Case-based reasoning (CBR) simulates the cognitive process of humanity, enjoyinga strong theoretical background and varied applications and has potential to solvethese problems.A traditional CBR credit scoring system is built to analyzeadavantages and disadvantages of the CBR when it is used in credit scoring. Thedisadvantages are mainly in the progress of CBR cycle. To solve these problems, anormalization algorithm for multiple attribute index, a hybrid multiple distancemeasurement algorithm and a calculation method with variable feature areintroduced in the case retrieval to optimize it and improve robustness of model.What`s more, Bayesian algorithm is employed as the reusing method to use the validdata fully. The results of an experiment shows that the improved CBR cycle systemnot only has a better performance on solving problems of biased sample anddynamic changing of credit samples; the optimized CBR cycle system is moreaccurate and has obvious progress in stability and explanatory function comparingto traditional system, but also has a better control on misclassification cost.