AIF-C01 Practice Question: A healthcare AI system predicts patient diagnoses

A healthcare AI system predicts patient diagnoses. The data collection process primarily samples from urban hospitals, leading to underrepresentation of rural populations. Which type of bias is this, and what is the most effective mitigation strategy?

⚠ Common exam trap

AIF-C01 often tests the distinction between bias types (representation vs. historical vs. measurement vs. aggregation), so candidates who see 'underrepresentation' but pick a mitigation that only reweights existing data choose the wrong answer.

Answer choices

Why each option matters

Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.

Correct answer & explanation

✓

Representation bias; collect additional data from rural hospitals

This is representation bias, which occurs when the training data underrepresents certain subgroups (rural populations) relative to the population the model will serve. The most effective mitigation is to collect additional data from the underrepresented group (rural hospitals) so the model learns their patterns and does not systematically misdiagnose them.

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

✓
Representation bias; collect additional data from rural hospitals
Why this is correct
Sampling only urban hospitals underrepresents rural populations, producing representation bias in the training data. Collecting additional data from rural hospitals rebalances the dataset so the model learns patterns from the missing subgroup, directly addressing the sampling gap rather than adjusting outputs after training.
✗
Historical bias; reweight urban samples to reduce their influence
Why it's wrong here
Sampling only urban hospitals underrepresents rural populations, which is representation or sampling bias, not historical bias. Reweighting urban samples is tempting because reweighting genuinely corrects skewed class distributions, but it cannot manufacture the missing rural records.
✗
Aggregation bias; use regularization to simplify the model
Why it's wrong here
Sampling only urban hospitals is representation bias, not aggregation bias, which concerns mixing distinct subgroups into one model. Regularisation addresses variance, not missing rural data. Aggregation bias would apply where one model is fitted across heterogeneous groups that need separate models.
✗
Measurement bias; apply data augmentation to rural records
Why it's wrong here
The flaw lies in who was sampled, not in how a variable was measured, so it is not measurement bias. Data augmentation is tempting because it genuinely helps when a measured feature is systematically distorted, but synthesising rural records cannot replace genuine rural data.

Go deeper

AIF-C01 Practice Question: A healthcare AI system predicts patient diagnoses

Why each option matters

Option-by-option breakdown

Related to this question