Society of Actuaries (SOA) PA Practice Exam

What is one advantage of using bagging in model building?

1. It increases model bias

2. It reduces variance without affecting bias

3. It simplifies the model interpretation

4. It only handles numerical variables

The correct answer is: It reduces variance without affecting bias

Using bagging, which stands for Bootstrap Aggregating, effectively reduces variance in model predictions while maintaining the same bias level. The technique involves creating multiple subsets of a training dataset by sampling with replacement. Each of these subsets is used to train individual models, and the final output is usually obtained by averaging the predictions (for regression tasks) or taking a majority vote (for classification tasks). The key benefit here is that because the models are trained on different subsets, they will likely produce different predictions for a given input. When these predictions are aggregated, the variability (or variance) of the model's predictions is reduced. This leads to more stable and reliable outcomes compared to a single model trained on the entirety of the data, which can be overly sensitive to noise or outliers. In contrast, while bagging does improve predictions, it does not inherently increase model bias; instead, it maintains the bias of the individual base models. Also, bagging does not simplify model interpretation significantly—especially in complex ensemble methods— nor is it limited to handling only numerical variables; it can work with categorical data as well. Overall, the main advantage of using bagging lies in its ability to reduce variance without a corresponding effect on bias, enhancing the robustness of the model's