Society of Actuaries (SOA) PA Practice Exam

Which of the following is a disadvantage of bagging?

1. It increases interpretability

2. It reduces variance

3. It retains advantages of decision trees

4. It can lead to loss of interpretability

The correct answer is: It can lead to loss of interpretability

The correct answer highlights a significant aspect of the bagging technique in statistical learning and machine learning. Bagging, or bootstrap aggregating, involves creating multiple versions of a dataset through resampling and fitting models to each version. The predictions from these models are then combined, typically by averaging for regression or majority voting for classification. One of the key advantages of bagging is that it improves predictive accuracy and reduces variance. However, while doing so, it can lead to a loss of interpretability. When using a single decision tree, the model is relatively easy to interpret since one can visualize and understand the decision-making process. In contrast, bagging combines multiple decision trees into an ensemble, making it significantly more complex. This complexity arises because the final predictions are based on the aggregated outcomes of many models rather than a single straightforward decision path. Thus, while bagging enhances prediction performance, it obscures the individual contributions of each model, making it harder to extract insights or understand the 'why' behind a given prediction. This makes the answer reflecting the potential loss of interpretability the correct choice.