Skip to Content

Overfitting & Underfitting

Start writing here...

Great question! Overfitting and underfitting are two common problems in machine learning that affect how well a model performs, especially on new, unseen data.

🔴 Overfitting

Definition: When a model learns the training data too well, including its noise and random fluctuations, leading to poor performance on new data.

  • Signs:
    • High accuracy on training data
    • Low accuracy on validation/test data
  • Causes:
    • Model is too complex (e.g., too many parameters)
    • Too little training data
    • Noisy data
  • Solutions:
    • Simplify the model
    • Use more training data
    • Apply regularization (like L1 or L2)
    • Use dropout (for neural networks)
    • Cross-validation

🔵 Underfitting

Definition: When a model is too simple to capture the underlying structure of the data, resulting in poor performance even on the training data.

  • Signs:
    • Low accuracy on both training and test data
  • Causes:
    • Model is too simple
    • Features are not informative
    • Insufficient training
  • Solutions:
    • Increase model complexity
    • Add better features
    • Train for longer
    • Reduce regularization

Quick Visual Metaphor:

  • Overfitting: Memorizing answers without understanding concepts.
  • Underfitting: Not learning enough to even memorize basic facts.

Want a quick graph that shows this visually (like a U-shaped curve of model complexity vs error)?