MLA-C01 Question 92
Select 2You are working on a machine learning classification model using a dataset with multiple numeric, text, and image features. During the exploratory data analysis, you notice that some classes are significantly underrepresented. To assess potential pre-training bias in the dataset, which metrics should you calculate?
- A
Class Imbalance (CI)
- B
Difference in Proportions of Labels (DPL)
- C
Confusion Matrix
- D
Mean Absolute Error (MAE)
- E
Per-Class Precision and Recall
Show answer and explanation
Correct answers: A, B
Explanation
Class Imbalance (CI) and Difference in Proportions of Labels (DPL) are both pre-training bias metrics that help identify imbalances in the dataset before training. CI quantifies the distribution of class labels, while DPL measures the differences in label proportions across groups. These tools are essential for ensuring that the dataset is fair and does not disadvantage certain classes or groups before training the machine learning model.
- A. Correct.
Class Imbalance (CI) is a key metric for identifying pre-training bias in datasets with underrepresented classes. It quantifies the imbalance in the distribution of labels across classes.
- B. Correct.
Difference in Proportions of Labels (DPL) is another important metric for assessing pre-training bias. It measures the difference in the proportions of labels between groups, which can highlight imbalances.
- C. Incorrect.
A confusion matrix is a post-training evaluation metric. It is not used to analyze pre-training bias but rather to evaluate model performance on a test dataset.
- D. Incorrect.
Mean Absolute Error (MAE) is a regression metric and is not applicable to evaluating pre-training bias in classification datasets.
- E. Incorrect.
Per-Class Precision and Recall are post-training metrics used to evaluate the performance of a trained model. They are not used for identifying pre-training bias in the dataset.