The telecommunications industry is confronted more and more to aggressive marketing campaigns from competitor carriers. Therefore, they need to improve the subscriber targeting to propose more attractive offers for gaining new subscribers. In the present effort, a five steps methodology to find new potential subscribers using supervised learning techniques over imbalanced datasets is proposed. The proposed technique applies community detection to infers consumption information of competitors carriers subscribers within the communities. Besides, it uses a sampling technique to reduce the effect of a dominant class for an imbalanced classification task. The proposal is evaluated with a real dataset from a Peruvian carrier. The dataset contains one-month data, which is about 200 millions of transaction. The results show that the proposed technique is able to identify between two to ten times more new potential clients, depending on the sampling technique, as shows using the top decile lift value.