Traditional machine learning models usually fail to perform well on the smaller or minority classes, which leads to unbalanced and biased predictions. To overcome this limitation, we propose an ensemble-based technique that combines the results of several models, each trained on a balanced subset of the main dataset. This technique makes use of the strengths of multiple models and can handle different kinds of missing data patterns such as MCAR, MAR, and MNAR. The effectiveness of the proposed approach is tested by checking its accuracy and comparing it with a Decision Tree model taken as the reference method. The results clearly show that the ensemble approach gives more consistent and accurate results for datasets that have class imbalance.