The approach
The task at hand is to find which features—whether quiz answers, technical, and demographic details—signal birth order the most, and if they are reliable to use for this purpose. One way to do this is by creating classification models to predict birth order, and then doing the following:
- Using the model's intrinsic parameters to discover which features impact the model the most. This concept is called feature importance, and it's a global modular interpretation method. This was explained in Chapter 2, Key Concepts of Interpretability, but we will go into more detail in this chapter.
- Exploring feature importance further with a more reliable permutation-based method called PFI.
- Examining the marginal impact to the outcome of the most important features with PDPs. That way, we can tell which feature values correlate the most with the predictions.
- Getting a more granular visualization of how individual features impact the models...
