Underfitting

Underfitting is a condition in which a machine learning model has not sufficiently learned from the training data and can only produce low-accuracy predictions. It occurs when a model is too simple to capture the important patterns and structures in the data, resulting in poor performance on both training and test data. For example, attempting to predict data with a complex nonlinear relationship using a simple linear model (a straight line) will result in underfitting. This is also described as "insufficient learning" or "excessive simplification" — it tends to occur when a model lacks the expressive capacity for the task at hand. Common causes of underfitting: • The model is too simple (e.g., using linear regression to predict a complex relationship) • Insufficient features, or missing important information • Inadequate training (e.g., too few epochs) • Overly strong regularization constraining the model's degrees of freedom Strategies to address underfitting: • Use a more complex model (e.g., a neural network) • Add more features or apply more sophisticated feature engineering • Train for longer (increase the number of epochs) • Relax the strength of regularization Underfitting represents a state where a model's performance potential is unrealized, and is the counterpart to overfitting. In machine learning, avoiding both extremes and finding the right balance — optimizing generalization performance — is essential.