AbstractAs a natural consequence of the multibillion dollar annual losses incurred as a result of credit card fraud, banks attach great importance to credit card fraud detection. In this paper, we proposed the use of six known models, namely, decision tree, random forest, Bayesian network, Naïve Bayes, support vector machine, and K* models, to form an ensemble for the detection of credit card fraud. We focused on the voting mechanisms used by the ensemble and proposed optimistic, pessimistic, and weighted voting strategies. The proposed model is called optimistic, pessimistic, and weighted voting in an ensemble of models. A dataset of real credit card transactions from a leading bank in Turkey was used to evaluate the performance of optimistic, pessimistic, and weighted voting in an ensemble of models. The results showed that the optimistic voting strategy enables the detection of 31.59% of fraudulent transactions with a false alarm rate of only 0.10%, the pessimistic voting strategy detects 93.92% of fraudulent transactions with a false alarm rate of 13.72%, and the weighted voting strategy detects 64.02% of fraudulent transactions with a false alarm rate of 0.75%. Banks can choose among these voting mechanisms depending on their preferred strategies for fraud detection and desired false alarm rates.