AI-102 Question 148
Select 4You are a data scientist tasked with evaluating the performance of a custom vision classification model deployed in Azure. The model classifies images into three categories: Cat, Dog, and Bird. During evaluation, you notice that the model performs well on classifying Cats and Birds, but struggles with Dogs, often misclassifying them as Cats. Which metric(s) should you prioritize to assess and address this issue?
- A
Precision for the Dog class
- B
Recall for the Dog class
- C
Overall accuracy of the model
- D
F1 score for the Dog class
- E
Confusion matrix for the model
Show answer and explanation
Correct answers: A, B, D, E
Explanation
In this scenario, the focus is on analyzing the performance of the model specifically for the Dog class, as it is underperforming compared to the other classes. Metrics such as precision, recall, and F1 score at the class level provide detailed insights into the model's ability to correctly identify and classify Dogs. The confusion matrix further supports this analysis by showing where misclassifications happen. Overall accuracy, while providing a high-level view, does not help isolate issues with specific classes and is therefore less relevant.
- A. Correct.
Precision for the Dog class is important because it measures how many of the images predicted as 'Dog' are actually 'Dog.' This helps identify if the model is overpredicting Dogs as another class, such as Cat.
- B. Correct.
Recall for the Dog class is crucial because it measures how many actual 'Dog' images are correctly identified by the model. Low recall indicates the model is missing many Dogs.
- C. Incorrect.
Overall accuracy of the model is not as useful in this scenario since it aggregates performance across all classes, potentially masking issues specific to the Dog class.
- D. Correct.
F1 score for the Dog class combines precision and recall into a single metric, making it valuable for understanding the model's performance specifically for the Dog class.
- E. Correct.
The confusion matrix provides a detailed breakdown of predictions versus actual labels for all classes, helping identify where misclassifications occur, such as Dogs being misclassified as Cats.