Malware detection is among the most extensively developed areas for computer security. Unauthorized, malicious software can cause expensive damage to both private users and companies. It can destroy the computer, breach the privacy of user and result in loss of valuable data. The amount of data uploaded and downloaded each day makes almost impossible for manual screening of each incoming software package. That is why there is a need for effective intelligent filters, that can automatically dichotomize between the safe and dangerous applications. The number of malware programs, that are faced by the detection system, is typically much smaller than the number of desired programs. Therefore, we have to deal with the imbalanced classification problem, in which standard classification algorithms tend to fail. In this paper, we present a novel ensemble, based on cost-sensitive decision trees. Individual classifiers are constructed according to an established cost matrix and trained on random feature subspaces to ensure, that they are mutually complementary. Instead of using a fixed cost matrix we derive its parameters via ROC analysis. An evolutionary algorithm is being applied for simultaneous classifier selection and assignment of committee member weights for the fusion process. Experimental analysis, carried out on a large malware dataset, prove that our method is capable of outperforming other state-of-the-art algorithms, and hence is an effective approach for the problem of imbalanced malware detection.