How to Fix AI Model Accuracy Disparities for Underrepresented Groups
A healthcare organization is deploying an AI model to predict patient readmission risk. The model was trained on historical data that underrepresented minority populations. During testing, the model shows lower accuracy for those groups. What should the data scientist do first?
Quick Answer
The correct first step is to re-evaluate the training data to ensure balanced representation and consider re-sampling techniques. This is because accuracy disparities for underrepresented groups typically stem from a skewed class distribution in the historical data, where the model learns patterns from the majority group while failing to generalize to minority populations. By auditing the data and applying re-sampling methods like oversampling the minority class or undersampling the majority, you directly address the root cause of the bias before altering the model architecture or features. On the Salesforce AI Associate exam, this scenario tests your understanding of the data-centric approach to fairness, often appearing as a trap where candidates mistakenly jump to feature engineering or algorithm changes. A common memory tip is “data first, model second”—always audit the training data for representational imbalance as the initial diagnostic step when you encounter lower accuracy for underrepresented groups.
⚠ Common exam trap
Salesforce often tests the misconception that removing sensitive attributes or improving model complexity automatically fixes bias, when in fact the data imbalance must be addressed first at the dataset level.
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
✓
Re-evaluate the training data to ensure balanced representation and consider re-sampling techniques.
The first step in addressing model bias is to audit the training data for representational imbalance. Re-evaluating the data and applying re-sampling techniques (e.g., oversampling minority groups or undersampling the majority) directly targets the root cause of the disparity—skewed class distributions—before modifying the model or its features.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Remove sensitive attributes like race and gender from the training data.
Why it's wrong here
Removing attributes does not eliminate bias from correlated features.
- ✗
Ignore the disparity because the model's overall accuracy is acceptable.
Why it's wrong here
Fairness requires equitable performance across groups.
- ✗
Retrain the model with more complex algorithms to improve accuracy.
Why it's wrong here
Complex algorithms may amplify bias if data issues persist.
- ✓
Re-evaluate the training data to ensure balanced representation and consider re-sampling techniques.
Why this is correct
Ensuring data representativeness addresses root cause of bias.
Go deeper
Related to this question
About these practice questions
Courseiva writes every AI Associate question from scratch — 753 in total, each with an explanation and a wrong-answer breakdown. None are copied from real exams or dumps. Learn why practice questions differ from exam dumps →
Same concept, more angles
1 more way this is tested on AI Associate
These questions test the same concept from different angles. Work through them to make sure you can recognise it however the exam phrases it.
Variation 1. During model development, the data scientist realizes the training data is not representative of the intended population. What should they do?
medium- A.Remove the underrepresented groups from the scope.
- B.Increase model regularization.
- C.Use the data as is, as the model will generalize.
- ✓ D.Augment data with synthetic samples for underrepresented groups.
Why D: Augmenting with synthetic data for underrepresented groups helps create a more representative dataset, reducing bias. Option A is wrong because removing groups from scope can lead to exclusion and increased bias. Option B is wrong because increasing model regularization does not address representativeness. Option C is wrong because using non-representative data as is can result in biased models that do not generalize well.
JA
Written by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
This AI Associate practice question is part of Courseiva's free Salesforce certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the AI Associate exam.