MLA-C01 Question 90
Select 2You are tasked with building a machine learning model to classify images of cats and dogs. During the exploratory data analysis phase, you notice that the dataset contains 75% images of dogs and 25% images of cats. Which pre-training bias metric(s) could you use to quantify this imbalance and assess potential bias in the dataset?
- A
Class Imbalance (CI)
- B
Difference in Proportions of Labels (DPL)
- C
Shannon Entropy
- D
Confusion Matrix
- E
Gini Impurity
Show answer and explanation
Correct answers: A, B
Explanation
Class Imbalance (CI) and Difference in Proportions of Labels (DPL) are both valid metrics for quantifying and assessing pre-training bias in datasets with imbalanced classes. CI measures the unequal representation of classes, while DPL compares the proportions of labels. These metrics are critical for identifying potential bias before training the model, especially in datasets like this one, where there is a significant imbalance in the number of images for each class.
- A. Correct.
Class Imbalance (CI) is a common metric used to evaluate the distribution of classes in a dataset. It would help quantify the unequal representation of cats and dogs in this case.
- B. Correct.
Difference in Proportions of Labels (DPL) is a metric that compares the proportions of different labels in a dataset. It would be relevant for assessing the imbalance in the proportions of cats and dogs.
- C. Incorrect.
Shannon Entropy is a measure of uncertainty or diversity in the dataset, but it is not specifically used to evaluate class imbalance or label proportions.
- D. Incorrect.
A Confusion Matrix is used to evaluate the performance of a trained model, not to assess pre-training bias in the dataset.
- E. Incorrect.
Gini Impurity is primarily used in decision tree algorithms to measure the impurity of a split, not for pre-training bias assessment.