A closer look at voting methods for cost-sensitive ensembles

Cost-sensitive prediction models have emerged as an alternative in scenarios where different types of prediction errors bear different costs.
For example, incorrectly predicting a fraudulent credit card transaction as legitimate is more costly than the other way around. Instead of just looking at a single model, Ensembles such as Bagging, AdaBoost, Random Forests, and their cost-sensitive variants, combine the outcome of several models in hope to get a more accurate prediction. The aim of this work is to closely investigate all possible ways of doing so, also known as ensemble voting, and compare their performance using large and imbalanced datasets.