Multi-objective Evolutionary Clust-Ering Algorithm With Its Applications In Credit Risk Management

The rapid development of China's financial industry and the explosive growth of financial data have led to the exploration of data mining technology in the financial sector.Especially in the field of credit risk management,while the traditional expert system method has high subjectivity and the general discriminant method is difficult to cope with increasingly complex high-dimensional data,various data mining technologies have shown the advantages of objectivity and accuracy.Cluster analysis is one of the core technologies of unsupervised data mining.In the absence of training samples,the characteristics of various samples in the cross-sectional aspect and time series aspect can be identified through the internal links of the data.So for the credit risk assessment problem and the credit risk linkage analysis problem,clustering analysis shows strong applicability.However,the existing clustering algorithms mostly only perform single clustering objective function,showing poor algorithm robustness and they are easy to fall into local optimum when dealing with high-dimensional credit risk data.Aiming at these problems,this thesis proposes a clustering algorithm for highdimensional data of credit risk,and applies it to the study of credit risk problems of two types of data: cross-section data and time series data.Specifically the following work is carried out:Reference Vector-based Multi-Objective Clustering(RVMOC)is proposed.Based on the idea of subspace clustering,we establish a multi-objective optimization problem for high-dimensional data clustering of credit risk and solve it by evolutionary algorithm.We also design a reference vector based local search method in the algorithm to improve the accuracy of the algorithm.Furthermore,a knee-pruning clustering ensemble for final solution selection method is proposed to obtain the final clustering result from the last generation solution set.Along with other comparison algorithms,the proposed RVMOC is tested on benchmark datasets.Credit Risk Application Study 1: Research on credit risk assessment of listed companies orienting cross-section data.We construct the credit risk index evaluation system and the credit risk criteria of listed companies.The proposed RVMOC algorithm is used to cluster the financial data of listed companies to determine whether the sample company has high credit risk and identify the key factors affecting credit risk.Credit Risk Application Study 2: Research on credit risk linkage of securities companies orienting time series.We employ CCA model to measure the credit risk of Chinese listed securities companies,and then the time series result data is obtained.Then we apply the proposed RVMOC algorithm to clustering the complete time series and sub-time series,and identify the credit risk characteristics of different securities companies,and discuss the relationship between risk linkage and systemic credit risk.The results show that the RVMOC algorithm proposed in this thesis has high accuracy and robustness in dealing with high-dimensional data clustering,and it shows good applicability in the practical application of credit risk.For the application research on credit risk assessment,we conclude that when evaluating the credit risk of listed companies,the financial data of the past two years contributes to more accurate results.Two years ahead of the profitability indicator and the one-year-ahead profitability,solvency and growth ability indicators have important reference significance.For the research on credit risk linkage of securities companies,the results show that the strengthening of the risk linkage of securities companies will lead to the accumulation of systemic credit risks,trigger large-scale risk outbreaks.When the degree of risk linkage declines,the resistance of securities company systems to risks can be improved,then the risk is gradually absorbed internally and eventually returns to normal levels.