You're training a supervised learning model and notice that training performance is strong, but validation performance is much weaker. You need to improve generalization without losing too much signal.
How would you handle overfitting in a predictive model?