Credit Evaluation Based on Spectral Clustering Method

The credit reports of the People's Bank of China consists of a mixed class of both numerical and nominal data. How to distinguish users' credit with their characteristics is the bottleneck of using the reports. This paper proposes a clustering method based on construction a weighted distance of users' mixed attributes, which is called spectral clustering. Then we take 1,198 records of credit data with 16 nominal attributes and 14 numerical attributes to conduct the experiment, and form several experiments, it can be seen that spectral clustering has good results and strong interpretability.