How would you approach a situation where your machine learning model is underperforming?