When would I use a decision tree instead of a PWE model?

PWE models perform a similar function to a Decision Tree although both have advantages and disadvantages. Two high level differences are:

• The Decision Tree is able to produce rules that are based on interactions between the characteristics (for example, young males behaving differently to young females).

  A downside to this is that if there are not interactions in the data, you are not gaining this benefit of the Decision Tree, and more data will be required to create an equivalent model with a Decision Tree as say using the PWE method.

• PWE models are good where you have characteristics that are good predictors but are available only for some customers. For example, you may only have SIC codes for some B2B customers, but you may have another classification variable for others.

  A downside to this is that a PWE model will be biased if it is built using correlated variables (which have good coverage in the data). A Decision Tree is not susceptible to bias in this way.

Further reading¶

There is a more detailed discussion of this topic in the Apteco white paper “Comparing PWE With Decision Tree and Other Model Building Techniques”.