RAS question
In machine learning, 'overfitting' refers to:
Correct answer: (C) A model that learns training data too well, including noise, and performs poorly on unseen data.
In machine learning, overfitting is when a model learns the training data too closely, including noise, and therefore performs poorly on new or unseen data.
Explanation
Overfitting is a failure of generalisation, not a sign that the model has become genuinely smarter. Option C is right because IBM defines overfitting as a model fitting too closely, or even exactly, to its training data, leaving it unable to make accurate predictions from other data. The usual reason is that the model has started memorising irrelevant noise or random detail instead of the dominant pattern. That is why an overfitted model can show low error or high accuracy on training data but perform badly when tested on new data. Cross-validation helps detect the gap, while regularisation and dropout are methods used to reduce noise and variance.
Why the other options are wrong
- (A) Using too little training data can contribute to overfitting, but overfitting is defined by fitting the training data too closely and failing to generalise, not by dataset size alone.
- (B) A model that fails to learn any pattern is underfitting, whereas overfitting occurs after the model has fitted the training data too closely.
- (D) Good performance on both training and test data indicates generalisation, while overfitting is signalled by strong training performance and weak performance on new or test data.
Concept
This tests the Science and Technology concept of machine-learning model training and generalisation. It recurs in RAS-style questions because overfitting is a core way to distinguish memorising training data from learning a usable pattern.