Abstract:A method combined of F-scores and support vector machine for customer classification was proposed, which can overcome the shortages of the existing customer classification method such as strict hypothesis, poor generalization ability, low prediction accuracy and low learning rate etc., and was applied to the problem of bank credit card customer classification. Empirical results show the validation accuracies of the final model can achieve 95% or more, which concludes that learning and generalization abilities of this model are excellent.