Nowadays, Telemarketing is an interactive technique of direct marketing that
many banks apply to present a long term deposit to bank customers via the phone.
Although the offering like this manner is powerful, it may make the customers
annoyed. The data prediction is a popular task in data mining because it can be
applied to solve this problem. However, the predictive performance may be
decreased in case of the input data have many features like the bank customer
information. In this paper, we focus on how to reduce the feature of input data and
balance the training set for the predictive model to help the bank to increase the
prediction rate. In the system performance evaluation, all accuracy rates of each
predictive model based on the proposed approach compared with the original
predictive model based on the truth positive and receiver operating characteristic
measurement show the high performance in which the smaller number of features.