Occurs when the training data do not reflect the target population.

Occurs when the training data do not reflect the target population.

• A. Imbalanced Dataset
• B. Representation Bias
• C. Explainability
• D. Transparency

Answer: (B)

When the data used to train a model don’t reflect the real population it will be applied to, the model learns patterns that fit the sample rather than the true population. This mismatch creates representation bias, where predictions or decisions are systematically biased against underrepresented groups or scenarios. In risk contexts, that means the model may underperform for certain segments, leading to inaccurate risk estimates and potential fairness or regulatory concerns. Mitigation focuses on improving how the data represent the target population—collect more representative examples, use sampling or weighting to balance the distribution, or apply methods that adjust for known disparities. Imbalanced datasets describe uneven counts of outcomes within the training data (for example, many more non-defaults than defaults). That can harm performance but isn’t the same as the data failing to reflect the whole target population. Explainability and transparency relate to understanding and revealing how the model works or its internals, not to whether the training data represent the population.

When the data used to train a model don’t reflect the real population it will be applied to, the model learns patterns that fit the sample rather than the true population. This mismatch creates representation bias, where predictions or decisions are systematically biased against underrepresented groups or scenarios. In risk contexts, that means the model may underperform for certain segments, leading to inaccurate risk estimates and potential fairness or regulatory concerns. Mitigation focuses on improving how the data represent the target population—collect more representative examples, use sampling or weighting to balance the distribution, or apply methods that adjust for known disparities.

Imbalanced datasets describe uneven counts of outcomes within the training data (for example, many more non-defaults than defaults). That can harm performance but isn’t the same as the data failing to reflect the whole target population. Explainability and transparency relate to understanding and revealing how the model works or its internals, not to whether the training data represent the population.