Nowadays, direct marketing is widely used advertisement method by many business areas such as banks in recent years. It is main purpose is to maximize return on investment, minimize cost of promotions and reach to peak number of customers that prefers the offerred campaign. Therefore, it is necessary to collect and process huge amount of customer related data to decide questions of which customer will be offered a product, which product will be suitable to him/her and via which channel the promotion will be presented. However, because positive customer response rates are much less than negative ones, negative data instances dominate positive ones and cause imbalance in dataset. This problem makes it difficult to make a successful selection of product and channel for a promotion and therefore, brings about decrease on true predictions and total profit value while false predictions and total cost value increase. In this thesis, methods are proposed which improve profit/cost ratio to increase return on investment while increasing accuracy rate. Experiments with proposed methods applied on a real bank dataset show very promising profit/cost ratios and accuracy rates on predicting customers with proper products and channels. Results of experiments indicate that proposed methods yields some amount of decrease on total profit value; however, since the decrease rate of total cost value is much greater than total profit one, profit/cost ratio increases.