Course contentsShow
← Machine Learning and Deep Learning
Lesson 143 of 3,5382. Machine Learning Mindset and Core ConceptsFree lesson

Overfitting vs Underfitting Recognition

Diagnosing model performance issues through training and validation error analysis.

Overfitting vs Underfitting Recognition

What you'll learn: How to diagnose whether your model is overfitting or underfitting by analyzing training and validation errors.

The Diagnostic Framework

Remember the bias-variance tradeoff? Now you'll learn to see it in action. Overfitting and underfitting reveal themselves through a simple pattern: comparing how your model performs on training data versus validation data.

Think of it like studying for an exam. If you memorize answers word-for-word (overfitting), you ace practice tests but bomb the real exam. If you barely study (underfitting), you perform poorly on both.

Reading the Error Signals

Underfitting (High Bias):

  • Training error: High
  • Validation error: High (similar to training error)
  • What it means: Your model is too simple to capture the underlying patterns
  • Example: Using a straight line to fit a curved relationship

Overfitting (High Variance):

  • Training error: Very low
  • Validation error: Much higher than training error
  • What it means: Your model memorized training data instead of learning generalizable patterns
  • Example: A complex model that wiggles through every single training point

Good Fit (Just Right):

  • Training error: Low
  • Validation error: Low and close to training error
  • What it means: Your model learned meaningful patterns that generalize well

The Gap Is the Tell

The gap between training and validation error is your key diagnostic tool. A large gap screams "overfitting!" Similar poor performance on both suggests "underfitting!"

Key Takeaway: Diagnose overfitting by looking for low training error but high validation error; diagnose underfitting when both errors are unacceptably high and similar.