AWS Certified AI PractitionerResponsible AIMedium
A healthcare provider is implementing an AI system to assist with disease diagnosis. During testing, it is discovered that the AI model consistently misdiagnoses patients from certain underrepresented ethnic groups due to a lack of sufficient training data for those groups. Which of the following is the most effective strategy to address this specific issue?
- ACollecting and augmenting data specifically for underrepresented groups.
- BIncreasing the overall size of the training dataset without specific focus.
- CImplementing a more complex neural network architecture.
- DReducing the number of features used in the model to simplify it.
Show answer & explanationAnswer & explanation
Correct answer: A. Collecting and augmenting data specifically for underrepresented groups.
The problem is explicitly stated as a lack of sufficient training data for underrepresented groups. The most direct and effective solution is to address this data imbalance by collecting more data for those specific groups or using data augmentation techniques.
Why the other options are wrong
- B. Increasing overall data size without focusing on the imbalance won't necessarily fix the underrepresentation issue.
- C. A more complex architecture might exacerbate the problem if the underlying data bias isn't addressed.
- D. Simplifying the model by reducing features might decrease its overall diagnostic accuracy and not solve the bias.
Addressing Data Bias
Addressing data bias involves identifying and mitigating unfair or skewed representation in the training data used for AI models.
- Data bias can lead to discriminatory or inaccurate AI outcomes.
- Strategies include data collection, augmentation, re-sampling, and re-weighting.
- Crucial for achieving fairness and equitable performance across groups.
Memory trick: To balance the data scales, you must add more to the lighter side.