A large retail bank wants to develop an ML system to help the risk management team decide on loan allocations for different demographics. What must the bank do to develop an unbiased ML model?

Question

Options

• AReduce the size of the training dataset.
• BEnsure that the ML model predictions are consistent with historical results.
• CCreate a different ML model for each demographic group.
• DMeasure class imbalance on the training dataset. Adapt the training process accordingly.

• A
  12% (4)
• B
  3% (1)
• C
  6% (2)
• D
  79% (26)

Explanation

Why D is Correct: Measuring class imbalance in the training dataset is essential because if certain demographic groups are underrepresented, the model will be biased toward the majority group, leading to unfair loan decisions. By identifying imbalance and adapting the training process (e.g., through resampling, reweighting, or using fairness-aware algorithms), the bank ensures the model learns equitably across all groups.

Why the Distractors Are Wrong:

• A is wrong because reducing the training dataset would worsen imbalance and reduce the model's ability to generalize - more representative data is better, not less.
• B is wrong because historical results often encode existing biases (e.g., past discriminatory lending practices), so replicating them would perpetuate unfairness rather than eliminate it.
• C is wrong because building separate models per demographic group can introduce inconsistency, increase complexity, and may itself constitute discriminatory treatment rather than fair, unified decision-making.

Memory Tip: Think of it as "Measure before you model" - bias in = bias out. If your training data doesn't fairly represent all groups, no amount of model tuning will fix the underlying problem. Always audit your data first.

No community discussion yet for this question.