关键词: 数据挖掘, Logistic回归, Logit变换, χ2检验, 积矩相关系数, 最大似然估计, 马氏距离, K-S统计量

Abstract: The illegal-cash-advance is one of the major fraud risks of the credit-card industry. There is almost no difference between single illegal-cash- advance transaction and the normal one, so it could not distinguish them based on different characteristics. To detect the illegal-cash-advance accounts automatically and accurately, the authors picked up correlative variables first, then made a business analysis, and took advantage of the nonlinear curve feature—the one defection of Logistic, and overcame the sensitivity of the multidimensional relativity between independent variables — the another defection, at last computed the weight of coefficient, and constructed a credit scoring model. The applications indicate that the accuracy of the model has achieved 82.72%.

Key words: data mining, Logistic regression, Logit transformation, χ2 test, product moment correlation coefficient, Maximum Likelihood Estimate (MLE), Mahalanobis distance, K-S statistic