What is overfitting in machine learning, and how can it be avoided?