AIF-C01 Question 179
Single answerAn e-commerce company is using a machine learning model to recommend products to users. During testing, the model performs well on training data but fails to recommend relevant products to users from certain demographic groups in real-world scenarios. What could be the most likely cause of this issue?
- A
The model is suffering from high variance, leading to overfitting on the training data.
- B
The model is suffering from high bias, leading to underfitting on the training data.
- C
The training data lacks diversity and is not representative of all demographic groups.
- D
The model uses a feature that is correlated with demographic groups, causing unintended bias in predictions.
Show answer and explanation
Correct answer: C
Explanation
The issue described in the scenario is most likely due to a lack of diversity in the training data, causing the model to perform poorly for certain demographic groups. Representativeness in training data is critical to ensure fair and accurate performance across all user groups. Without sufficient representation, the model may fail to generalize well to underrepresented groups, even if it performs well on the training data.
- A. Incorrect.
High variance typically causes the model to overfit the training data, but this does not directly explain the model's failure to perform well for certain demographic groups in real-world scenarios.
- B. Incorrect.
High bias generally results in underfitting and poor overall performance, but it does not specifically explain the disparities in performance across demographic groups.
- C. Correct.
If the training data lacks diversity or representation of certain demographic groups, the model may not learn patterns for those groups, leading to poor performance in real-world scenarios. This is the most likely cause of the issue described.
- D. Incorrect.
While feature correlation with demographic groups can introduce bias in some cases, the question does not provide specific details indicating that this is the issue here.