| The populari.ty of credit card has greatly facilitated the transactions between merchants and cardholders.However,a large number of fraudulent cases have arisen.Every year,billions of euros are lost due to credit card fraud.Therefore,financial institutions urgently need a well-designed fraud detection system to prevent fraudulent transactions.In recent years,machine learning technology has been widely used in credit card fraud detection and achieved favorable performances.However,the credit card fraud detection system needs to be constantly improved due to the highly imbalance and concept drift of credit card dataset.This thesis,research of scheme for credit card fraud detection based on machine learning,is aimed at solving the serious problem existed in the current fraud detection recommendation system.The proposed en semble learning framework based on training set partitioning and clustering not only ensures the integrity of the sample features,but also solves the highly imbalance of the dataset,it avoids the problem that under-sampling potentially discarding valuable samples.The proposed ensemble learning algorithm combining active learning and semi-supervised learning solves the problem of highly cost of labeling credit card data manually,it can use lots of relevant samples in time while saving a great deal of manpower cost.The proposed sample selection algorithm based on highly posterior probability and diversity for active learning solves the problem that there are many false alarms in the current fraud detection system,it enables financial institutions to use manpower to verify highly suspicious transactions that possess rich diversity.Then samples possessing rich diversity can strengthen the training and improve the performance of the fraud detection model.Finally,we conduct extensive experiments on a real dataset and a simulated dataset and we compare the proposed scheme with the state-of-the-art fraud detection models,the experimental results show the superiority of our proposed scheme. |
