Example: apartment-prices data

In this section, we use a random forest model to predict price per square meter for an apartment. Focus on two variables, surface and construction year

Partial-dependence-profiles

Source: Figure 17.5

Clustered partial-dependence profiles

Source: Figure 17.6

Grouped partial-dependence profiles

In this example, we use district as grouping variable to see if relationship between model’s prediction with construction year and surface is similar in different geographic areas.

 

Constrastive partial-dependence profiles

Below we compare the PDP output for from two predictive models: a basic linear regression model and the random forest model.

Source: Figure 17.8