Exams / AIF-C01 / Bias, variance and datasets / Question 1
AIF-C01 practice question: Bias, variance and datasets (1 of 2)
Guidelines for Responsible AI A free question from our original AIF-C01 practice exam. Try it, then open the answer to see why each option is right or wrong.
A model that predicts delivery delays scores 98 per cent accuracy on its training data but only 71 per cent on new data it has not seen. What is the MOST likely problem?
Choose one answer.
- AA. Underfitting, because the model is too simple to learn the patterns
- BB. Overfitting, because the model has memorised the training data
- CC. Data drift, because production data has changed since launch
- DD. Hallucination, because the model is inventing delivery data
Show the answer and explanation
Answer: B
- A. Wrong: An underfitting model performs poorly on its training data as well. This model does very well in training.
- B. Right: Correct. Strong training performance with weak results on unseen data is the sign of overfitting (high variance).
- C. Wrong: The gap shows up on held-out data at evaluation time, before any change in production data over time.
- D. Wrong: Hallucination describes generative models producing false content, not a gap between training and test accuracy.
Explanation
Overfitting (high variance) means the model fits the training data too closely and fails to generalise. Underfitting (high bias) means it is too simple and performs poorly even in training. Remedies for overfitting include more and more varied data, fewer features and more regularisation.
Study
Keep practising
Answer questions with a score at the end, or practise just this topic with the explanation after each one. No sign-up needed.