Underfitting occurs when a machine learning model is too simple to capture the underlying structure of the data, resulting in poor performance on both the training data and unseen data. This happens when the model is not complex enough to learn the patterns and relationships present in the data.

Key characteristics of underfitting include:

High Bias: Underfit models have high bias, meaning they make strong assumptions or simplifications about the data that may not reflect the true underlying relationships.

Poor Performance: Underfit models typically have high error rates on both the training data and unseen data. They fail to capture the nuances and complexities of the data, leading to inaccurate predictions or classifications.

Too Simple: Underfit models are often too simple or have too few parameters to adequately represent the data. They may lack the flexibility to capture non-linear relationships or complex patterns present in the data.

Underfitting can be addressed by increasing the complexity of the model, adding more features, or using more sophisticated algorithms. However, it's essential to strike a balance between model complexity and the amount of available data to avoid overfitting, where the model learns noise or random fluctuations in the data. Cross-validation and performance evaluation on validation data can help identify and mitigate underfitting in machine learning models.