| The economic growth and various measures on the liberalization and internationalization of finance, promoting the credit card market continues to grow up in both domestic and foreign countries. At the same time, credit card fraud transactions have gone up at a stupendous speed. How to detect credit card fraud transactions effectively, rapidly and exactly has been a commonly concerned problem in the financial industry at present.The imbalanced distribution of credit card fraud data sets makes the classification performance of classical data mining algorithms is dissatisfactory. Based on the research and analysis of the characteristics of credit card fraud data sets, this paper proposes to separate the major class data sample to two clusters by using K-Means cluster algorithm according to the characteristics of imbalanced set, therefore engender leaf nodes which meet certain conditions. Integrating leaf node into minority class data sets and form a new subset, will ensure this distribution achieve balance with the premise of the introduction of noise-free data and integrality of data. At the same time against the limitations of a single classifier, we introduce the theory of combined classifiers fusion, and receive an assembled classifier model based on the combination of AdaBoost and Cost-sensitive which C4.5 is the base classifier. The result of the experiment indicates above theory will improve the imbalanced distribution result of credit card data efficiently.It is obvious that the characteristics of class imbalanced distribution in credit card fraud and the exist of minority class makes the overall accuracy can not evaluate the classification results accurately . Based on Precision and Recall of minority class, this thesis will adopt F-Measure indicator to evaluate the results aimed at the classification results of minority class sample. The experimental result has proven to be better in line with the characteristics of classification of credit card. |
PDF Full Text Request