Research On Credit Card Fraud Detection Based On Machine Learning Algorithm

With the rapid development of the times and the advent of the digital economy era,people’s concepts of consumption and consumptive behaviors are constantly changing.More and more people tend to use credit cards for online consumption,which not only brings convenience to our lives,but also provides an opportunity for some criminals to implement credit card fraud.This phenomenon has caused huge economic losses to individuals and banks.Therefore,it is essential to strengthen the detection of credit card fraud.This thesis studies the problem of credit card fraud detection based on real online credit card transaction data.Firstly,we preprocessed the data by missing data imputation method and data normalization method,and used the combination of SOMTE and Tomek Links to solve the problem of data category imbalance.Then,124 important features were selected by using univariate feature selection method and recursive feature elimination method.After that,we built models based on the processed data set and compared the results of the models.The established models mainly include logistic regression model,random forest model,LightGBM model,XGBoost model and Stacking model.Finally,we concluded that the LightGBM model and the XGBoost model performed better than the other two single models in credit fraud detection.Compared with the other four single models,the Stacking model has a certain improvement.And the Stacking model can be better applied to credit card fraud detection.