Credit Card Approval Prediction by Non-negative Tensor Factorization

With the increasing number of credit card applications, banks are opting towards the use of prediction-based algorithms as opposed to manual approval methods. Data analysis has exhibited a strong correlation between several financial and personal factors of a client and the likelihood of said client complying with their respective bank's credit policies. In this paper, we propose the use of the PARAFAC tensor factorization method to predict and grant credit cards to applicants based on the customers' activity history. We used six financial and personal factors and constructed a tensor by reducing them into three factors. We predicted the resulting factors through the use of alternating least squares algorithm with an emphasis on error minimization and finally re-constructed the original tensor. Using this tensor factorization, the machine-learned which of these applicants are most likely to accumulate bad debts and granted or rejected the applications based on the prediction.