Society of Actuaries (SOA) PA Practice Exam

Do decision trees require variable transformations for numeric predictors?

1. Yes, because they are sensitive to data distributions

2. No, because they utilize relative order of values to make splits

3. Yes, to ensure normality of the data

4. No, but scaling is recommended

The correct answer is: No, because they utilize relative order of values to make splits

Decision trees do not require variable transformations for numeric predictors because they rely on the relative order of values to make splits. This characteristic allows decision trees to handle a wide range of data types without needing transformations that might be necessary for other modeling techniques, particularly those that assume linear relationships or normality of the predictors. In a decision tree, the algorithm makes decisions by determining the best way to split the data based on a chosen criterion, such as minimizing impurity or maximizing information gain. Since the splits are based on the relative ordering of the numeric values, it does not matter whether those values follow a specific distribution. Each predictor's thresholds are determined from the actual values present in the data, allowing the model to adaptively create rules that best separate the outcomes. This flexibility is a significant advantage of decision tree models, as they can work effectively with both continuous and categorical predictors without preprocessing steps like normality checks or scaling. Consequently, they can efficiently address real-world datasets that do not conform to standard assumptions.