All terms

Underfitting

When a model is too simple to capture the real patterns in the data — poor performance on both train and test.

Machine Learning1 min read

Definition

An underfit model fails to learn the signal — high error on training data and worse still on new data. Causes include too little capacity, too strong regularization, or insufficient training.

It is the opposite failure mode of overfitting.

In simple terms

Overfitting is memorizing the practice test. Underfitting is barely studying — you fail the practice test and the real exam.

Where you see it

  • A linear model on highly nonlinear data.
  • Stopping training too early.
  • Using tiny models for complex language tasks.

How it works

  1. 1.Diagnose

    Train and validation error are both high.

  2. 2.Increase capacity

    Richer features, deeper nets, or longer training.

  3. 3.Ease constraints

    Reduce excessive regularization if needed.

Why it matters

  • Recognizing underfit vs overfit tells you whether to simplify or enrich the model.

Often confused

  • If validation error is high, you must be overfitting.

    High train error points to underfitting; low train / high val points to overfitting.