Chapter 2: Interpretable Models
Some machine learning models are already inherently interpretable, e.g. simple LMs, GLMs, GAMs and rule-based models. These models are briefly summarized and their interpretation clarified.
-
Chapter 2.1: Inherently Interpretable Models - Motivation
In this section, we provide reasons supporting the choice of an inherently interpretable model as a first step. We show advantages and disadvantages and justify why both simple and complex models are worth considering.
-
Chapter 2.2: Linear Regression Model (LM)
The linear regression model is probably the oldest and most widely used standard technique. In addition to briefly outlining the basics of LMs, we show how this model is inherently interpretable.
-
Chapter 2.3: LM - Interactions and LASSO
Since the linear regression model in its standard form is not very flexible, this section reviews two of the most popular adjustment techniques: inclusion of higher order and interaction effects; and LASSO.
-
Chapter 2.4: Generalized Linear Models
This section deals with generalized linear models (GLM) by introducing common types of GLMs. A strong focus is on binary classification problems where logistic regression plays a major role.
-
Chapter 2.5: GAM & Boosting
A viable solution to problems with nonlinear relationships are Generalized Additive Models (GAMs), which can adapt quite flexibly to the data, but must be interpreted different from LMs/GLMs. Beside GAMs, the technique of boosting is presented in this section.
-
Chapter 2.6: Rule-based Models
Rule-based alorithms such as CART are intuitive up to a certain size. Therefore, Trees, RuleFit and Decision Rules belong to the group of inherently interpretable models.