A model scores 99% on its training data and 70% on new data. What is this called?
Options
- Underfitting
- Data drift
- Overfitting (Correct option)
- Class imbalance
Answer
Overfitting
Why
The model has memorized details and noise of the training set instead of learning the general pattern, so it does not carry over to new data.
Underfitting is the opposite: the score is poor on both training and new data.
Typical fixes: more data, a simpler model, regularization, or fewer features.
Remember
Great on training, poor on new data: overfitting.