Multi-decision-tree credit-card fraud detection method and system based on constraint projection

Abstract

The invention provides a multi-decision-tree credit-card fraud detection method based on constraint projection. The method includes: establishing a sample attribute set A, and acquiring a training sample set D; separating the training sample set D into a fraud-transaction sample set D<min> and a normal-transaction sample set D<maj>, and generating a must-link set set M={M<k>|k=1, 2, ..., K} and acannot-link set set C={C<k>|k=1, 2, ..., K} in an iterative manner; learning a projection matrix set in an iterative manner; generating a training set set after projection in an iterative manner; learning a decision tree set in an iterative manner; and processing a to-be-classified sample x, respectively projecting the x, predicting a category of the x, and finally determining whether the x belongs to normal samples or fraud samples. Compared with the related art, the multi-decision-tree credit-card fraud detection method based on constraint projection provided by the invention can effectivelyanalyze whether the transaction sample belongs to the fraud samples, can maintain high accuracy on predicting the normal samples and the fraud samples, and thus has a wider range of engineering application values. The invention also provides a multi-decision-tree credit-card fraud detection system based on constraint projection.

Description

technical field [0001] The invention relates to the technical field of data analysis, in particular to a credit card fraud detection method and system based on constraint projection with multiple decision trees. Background technique [0002] With the rapid development of my country's economy and information technology, the use of credit cards continues to rise, and has become an important medium in the payment field. The wide application of credit cards has brought convenience to people's life and work. However, the ensuing credit card fraud has also brought greater and greater economic losses. The problem of credit card fraud risk control has become a major factor affecting the further development of my country's credit card business. How to strengthen the identification and prevention of credit card fraud has also become an important issue in bank risk control. [0003] Faced with the current situation of credit card fraud, the main identification and prevention measures...

AI Technical Summary

Problems solved by technology

However, this high accuracy rate does not have practical significance for identifying fraudulent samples belonging to the minority class. Only the correct identification of fraudulent samples is the desired goal.

Traditional machine learning methods tend to build models with high accuracy, often ignoring the identification of minority samples

Images