Your machine learning model is not capturing the underlying patterns in the training data effectively, suggesting it's too simple. Which strategy, supported by Databricks MLlib, would you use to tackle this underfitting issue?