Underfitting occurs when a machine learning model is too simple to capture the underlying patterns in the data. This can happen when the model has insufficient complexity, such as using a linear model for a nonlinear relationship. As a result, the model performs poorly on both training and test datasets, failing to make accurate predictions. To address underfitting, one can increase the model's complexity by using more advanced algorithms or adding features to the dataset. Techniques like polynomial regression or neural networks can help improve the model's ability to learn from the data, leading to better performance and more accurate predictions.