| With the rapid development of online loans,consumer finance and other industries,the personal credit industry has shown unlimited vitality and potential.In order to ensure the healthy development of the credit business and maintain financial stability,how to accurately and efficiently assess personal credit ratings has become an urgent problem to be solved.Association rules are used to discover associations or correlations among a large number of data item sets,and decision fusion is the highest level of information fusion with good real-time and fault tolerance.Therefore,based on the theory of association rules and decision fusion,this article constructs the following three personal credit models:(1)Constructed a personal credit model integrating multi-view similarity measures(IMV-SM).The credit attribute of a certain credit rating can be regarded as a random vector obeying a certain distribution.Therefore,in this paper,the attributes of credit are considered as a whole.Under the assumption that the distribution of credit attributes of the same credit rating has the same mean value,The HotellingT2 statistic is selected to measure the similarity of customer credit;from the perspective of marginal distribution of each component of random vector,Jensen Shannon difference,an information measure based on information theory,is selected to measure the similarity and difference of customer credit;When the value of customer's credit attribute is regarded as the vector of high-dimensional data space,Cosine distance based on the angle between vectors is selected to measure the difference of customer's credit;finally,weighted voting strategy is used to fuse the decision information of the nearest neighbor classifier from three perspectives.(2)Building up a personal credit model hybridizing association rules and adaptive weighted decision fusion(HAR-AWDF).Apriori-based association rule mining is the process of obtaining strong association rules by mining frequent itemsets in a data set based on a given measure(support,confidence,or weighted chi-square).However,different attribute items,measures and rules have different ability to discriminate credit evaluation.Therefore,this paper uses the posterior probability settings to reveal the weight of the discriminative ability of the attribute items,sets the weight to reflect the credit contribution of the three measures through the classification performance setting,and sets the weight to reflect the rule evaluation ability with the aid of the learned threshold.Furthermore,mining out attribute items,frequent item sets and classification rules that can improve the credit evaluation performance more than traditional association rules complements the deficiencies of different measures.Finally,the weighted voting strategy is used to fuse the decision information of the classification rules.(3)Constructing a combination model of classification fusion using multi-view similarity measures and association rules(CFMSM-AR)for personal credit investigation.In addition to revealing the differences of customer credit from the three aspects of the parameter hypothesis test of distribution,the information content of distribution and the direction of vector,we can also regard the credit attribute value as a point in vector space,and quantify the similarity of customer credit risk through the measurement derived from the consistent norm in vector space Chebyshev distance;and when the credit attribute value is binary coded,we can also Hamming distance between equal length strings indicates the similarity of customer credit.But for high-dimensional data,Chebyshev distance,Hamming distance and Cosine distance are degraded because of"dimension disaster".Therefore,this paper uses association rules to extract important frequent attribute sets,in order to improve the performance of Chebyshev distance,Hamming distance and Cosine distance to identify customer credit risk.Finally,weighted voting strategy is used to fuse decision information of five similar measures. |
PDF Full Text Request