Visualizing Variable Importance and Variable Interaction Effects in Machine Learning Models

TL;DR

This paper introduces novel visualization techniques for interpreting variable importance, interactions, and partial dependence in machine learning models, enhancing understanding especially with many variables, and provides an R package implementation.

Contribution

It presents new, model-agnostic visualization methods for variable importance and interactions, including heatmaps, graph displays, and matrix layouts, applicable to regression and classification tasks.

Findings

New visualization techniques improve interpretability of models.

Methods are effective even with large numbers of variables.

The R package vivid facilitates practical application.

Abstract

Variable importance, interaction measures, and partial dependence plots are important summaries in the interpretation of statistical and machine learning models. In this paper we describe new visualization techniques for exploring these model summaries. We construct heatmap and graph-based displays showing variable importance and interaction jointly, which are carefully designed to highlight important aspects of the fit. We describe a new matrix-type layout showing all single and bivariate partial dependence plots, and an alternative layout based on graph Eulerians focusing on key subsets. Our new visualizations are model-agnostic and are applicable to regression and classification supervised learning settings. They enhance interpretation even in situations where the number of variables is large. Our R package vivid (variable importance and variable interaction displays) provides an implementation.