What is overfitting in machine learning?
A.
When a model fails to learn training data
B.
When a model learns training data too well and fails on new data.
C.
When the dataset is too small
D.
When the training takes too long
Correct answer is:
B. When a model learns training data too well and fails on new data.
Explanation:
The correct answer is When a model learns training data too well and fails on new data. because it describes a model that has memorized specific noise rather than learning general underlying patterns.
- Step 1 (Training Performance): An overfitted model fits the training dataset perfectly, achieving near-zero error margins by adapting to minor fluctuations and random noise within that specific data batch.
- Step 2 (Generalization Failure): Because it is tailored too narrowly to the training set, it fails to generalize accurately when presented with novel, unseen validation testing data.
- Incorrect Options:
- When a model fails to learn training data is incorrect because failing to learn the initial training dataset describes underfitting, not overfitting.
- When the dataset is too small is incorrect because a small dataset can cause overfitting, but it is not the definition of the concept itself.
- When the training takes too long is incorrect because lengthy training times relate to computational resource thresholds rather than structural generalization errors.