Society of Actuaries (SOA) PA Practice Exam

How does undersampling address the problem of unbalanced classes?

1. It increases the number of instances in the minority class

2. It keeps all instances of the majority class

3. It samples from both classes equally

4. It removes instances from the minority class

The correct answer is: It keeps all instances of the majority class

Undersampling is a technique specifically designed to address the problem of unbalanced classes in a dataset by modifying the composition of the majority class. This approach reduces the number of instances from the majority class so that it is more balanced with the minority class. By keeping all instances of the minority class intact while selectively removing instances from the majority class, undersampling helps to create a more equitable distribution of class labels. This adjustment can improve the performance of predictive models, as they are less likely to be biased towards the majority class. In contrast, other strategies, like oversampling (which increases the number of instances in the minority class), create a different dynamic, and maintaining all instances of the majority class does not inherently correct the class imbalance. Therefore, keeping all instances of the minority class while reducing those from the majority class is the fundamental principle of undersampling that effectively addresses class imbalance issues.