As a machine learning engineer, you are tasked with improving the accuracy of a text classification model that suffers from poor performance due to an unbalanced dataset. Which technique would be MOST effective in addressing this issue while maintaining the model's ability to generalize?