Research On Risk Identification Method Of Bank Guarantee Circle Based On Massive Data Analysis

With the development of credit market,the loan risk caused by personal loan or enterprise loan increases the amount of non-performing loans of banks year by year.By the end of 2018,the balance of non-performing loans has reached 2025.4 billion yuan.In order to reduce the occurrence of non-performing loans,banks mostly use mortgage,guarantees and other forms of loan business to offer loan transaction,and the guarantee network comes along with various risks.How to effectively identify the risk of bank guarantee circle based on massive data and timely stop loss is the focus of this thesis.Under normal circumstances,the guarantee network is relatively large,and the direct analysis is more complicated.Therefore,this thesis will study the risk identification method of the guarantee circle from the following three aspects:Firstly,through the depth-first traversal algorithm of the directed graph,each separate guarantee circle is identified from the complex guarantee relationship to realize the segmentation of the large guarantee network.The influence of the link degree and the link strength on node is comprehensively considered based on the PageRank algorithm such that key enterprises in the guarantee circle are discovered.Secondly,a FW-FS feature selection algorithm is proposed to preprocess the original credit data of the enterprise and select the most representative and non-redundant optimal feature subset covering all data or most of the original data information.At the same time,the problem of the K-value in the traditional feature selection algorithm is avoided when feature clustering.Thirdly,the key enterprises are investigated based on the SMOTE-Logistic regressive risk identification model of feature selection.And the high-risk enterprises within the guarantee circle are located and the domino effect brought by the risk of the guarantee circle is gradually reduced.At the same time,the problem of unbalanced distribution of original sample data is solved.Finally,the influence of data imbalance and FW-FS feature selection on the recognition results is verified and the presented method is compared with current popular algorithms including Decision Tree,AdaBoost and SVM.The experimental results show that the proposed method has a better performance in dealing with the risk identification problem of the guarantee circle.