Society of Actuaries (SOA) PA Practice Exam

What is a disadvantage of Generalized Linear Models (GLMs) when compared to Decision Trees?

1. GLMs automatically capture variable interactions

2. GLMs are always easier to interpret

3. GLMs do not capture the effects of variable interactions automatically

4. GLMs require complex decision trees for accuracy

The correct answer is: GLMs do not capture the effects of variable interactions automatically

The chosen answer is correct because Generalized Linear Models (GLMs) do not inherently include mechanisms to automatically capture interactions between variables. In GLMs, interactions must be explicitly defined and included in the model formulation by the analyst. This can be a limitation because the analyst must have prior knowledge or intuition about which variables might interact and must invest additional effort in model specification. In contrast, Decision Trees are designed to naturally account for interactions between variables as they split the data based on the values of the input features at each node. This allows them to capture different relationships within the data without requiring pre-specification of interactions. Other statements highlight aspects of GLMs that are not disadvantages; for example, GLMs do offer interpretability and can capture certain linear relationships well. However, their lack of automatic interaction handling is a distinct disadvantage when compared to Decision Trees, particularly in situations where interactions play a significant role in the data's structure.