Interpretable Subgroups and Interaction Detection

Abstract: Model-agnostic interpretation methods are often used in ML to produce explanations based on non-linear, non-parametric prediction models. Explanations are often represented in the form of summary statistics or visualizations, e.g., feature importance values or effects. Many interpretation methods describe the behavior of black-box models either locally for a specific observation or globally for the entire model and input space. Interpretable machine learning (IML) methods such as partial dependence plots visualize marginal feature effects but may lead to misleading interpretations when feature interactions are present. Hence, employing additional methods that can detect and measure the strength of interactions is paramount to better understand the inner workings of ML models. Furthermore, methods that produce regional explanations and lie between local and global explanations are rare and not well studied, but offer a flexible way to combine advantages between both types of explanations. Here, we will focus on subgroup approaches for IML methods, where interpretable areas in the input space are often induced by a combination of recursive partitioning and IML. Specifically, we will present regional effect plots with implicit interaction detection, a novel framework to detect interactions between a feature of interest and other features.
The framework also quantifies the strength of interactions and provides interpretable and distinct regions in which feature effects can be interpreted more reliably, as they are less confounded by interactions. In a second part, we will extend the REPID algorithm to our new framework GADGET (Generalized Additive Decomposition of Global Feature Effects), which is combinable with many existing effect estimation approaches, e.g. ALE or Shapley dependency plots.

Back