Overfitting and underfitting

After this lesson you can: Diagnose which of the two you have from train vs test performance.

First: Train/test split, and why it is non-negotiable

Watch

• Machine Learning Fundamentals: Bias and Variance - StatQuest with Josh Starmer, 7 min. Clear beginner-friendly bias-variance framing that directly explains why training perfection can fail on test data.
• Machine Learning-Bias And Variance In Depth Intuition| Overfitting Underfitting - Krish Naik, 17 min. Its title targets overfitting and underfitting directly, with enough duration for practical intuition.

Notes

Overfitting means the model learned the training examples too specifically, including their noise and accidents, so it looks excellent on the data it saw but fails on new data. Underfitting means the model is too simple, poorly trained, or missing useful signal, so it cannot even do well on the training data. The train-test split lets you see this difference because the training score measures how well the model fit what it saw, while the test score estimates how well it generalizes.

Unlock the full lesson notes, the exercises and the graded checkpoint - ₹5,000 once.