Imagine you are tasked with improving the accuracy of a model that is underperforming. What approach would you take?