5. Conclusion
As the external situation has undergone profound changes in operations, customer competition among
commercial banks is increasingly fierce. Customer attrition analysis has become an significant research topic for
commercial banks. This study aims to establish SVM model to predict customer attrition of commercial banks. Due
to the imbalanced characteristics of the actual commercial bank customer churn dataset, SVM model cannot predict
the churners effectively and only general evaluation criteria cannot measure the predictive power of the model.
By changing the sample distribution, Random sampling method has a higher degree of recognition. Therefore, we
use random sampling method to improve SVM method, and select F-measure to evaluate the predictive power. To
compare the prediction effect, we also establish Logistic regression model. The results show that the combination of
random sampling method and the SVM model can significantly improve the predictive power and help commercial
banks predict churners more accurately.