MLS-C01 · Question #366
A banking company provides financial products to customers around the world. A machine learning (ML) specialist collected transaction data from internal customers. The ML specialist split the…
The correct answer is B. Class imbalance (CI). The observation that a training dataset contains 'fewer examples of customers in the 40 to 55 year-old age group compared to the other age groups' indicates a disproportionate representation of a specific subgroup, which is classified as Class Imbalance (CI) pretraining bias in…
Question
A banking company provides financial products to customers around the world. A machine learning (ML) specialist collected transaction data from internal customers. The ML specialist split the dataset into training, testing, and validation datasets. The ML specialist analyzed the training dataset by using Amazon SageMaker Clarify. The analysis found that the training dataset contained fewer examples of customers in the 40 to 55 year-old age group compared to the other age groups. Which type of pretraining bias did the ML specialist observe in the training dataset?
Options
- ADifference in proportions of labels (DPL)
- BClass imbalance (CI)
- CConditional demographic disparity (CDD)
- DKolmogorov-Smirnov (KS)
How the community answered
(47 responses)- A4% (2)
- B77% (36)
- C6% (3)
- D13% (6)
Why each option
The observation that a training dataset contains 'fewer examples of customers in the 40 to 55 year-old age group compared to the other age groups' indicates a disproportionate representation of a specific subgroup, which is classified as Class Imbalance (CI) pretraining bias in Amazon SageMaker Clarify.
Difference in Proportions of Labels (DPL) measures the difference in the proportion of positive labels (e.g., target outcome) between different groups, not the raw count or representation of the groups themselves.
The observation that the training dataset contains 'fewer examples of customers in the 40 to 55 year-old age group compared to the other age groups' directly describes an unequal distribution or underrepresentation of a specific subgroup within the dataset. In Amazon SageMaker Clarify, this type of disparity in the count of observations for a particular facet (like an age group) is measured by the Class Imbalance (CI) metric for pretraining bias.
Conditional Demographic Disparity (CDD) measures bias based on conditional distributions, comparing outcomes or labels between groups given specific conditions, which is not what is described by a simple count disparity of a group.
Kolmogorov-Smirnov (KS) is a statistical test used to compare cumulative distribution functions, not a direct name for a type of bias describing unequal representation of groups.
Concept tested: SageMaker Clarify pre-training bias metrics
Source: https://docs.aws.amazon.com/sagemaker/latest/dg/clarify-considerations-bias-metrics.html#clarify-considerations-pre-training-bias-metrics
Topics
Community Discussion
No community discussion yet for this question.