Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DimVis: Interpreting Visual Clusters in Dimensionality Reduction With Explainable Boosting Machine

Parisa Salmanian, Angelos Chatzimparmpas, Ali Can Karaca, Rafael M. Martins

TL;DR

DimVis tackles the interpretability challenge of nonlinear dimensionality reduction by employing a contrastive, supervised Explainable Boosting Machine (EBM) that is trained in real time on user-selected data to explain clusters in DR projections. The approach yields single- and pairwise-feature importances visualized through accessible plots, enabling rapid, interpretable insight into why clusters form. Demonstrations on healthcare datasets (e.g., Breast Cancer Wisconsin, Pima Indian diabetes) illustrate the method’s ability to reveal domain-relevant feature contributions and uncertainties, supporting more trustworthy DR-based exploration. This tool has practical implications for analysts seeking transparent, interactive explanations of complex high-dimensional data visualizations.

Abstract

Dimensionality Reduction (DR) techniques such as t-SNE and UMAP are popular for transforming complex datasets into simpler visual representations. However, while effective in uncovering general dataset patterns, these methods may introduce artifacts and suffer from interpretability issues. This paper presents DimVis, a visualization tool that employs supervised Explainable Boosting Machine (EBM) models (trained on user-selected data of interest) as an interpretation assistant for DR projections. Our tool facilitates high-dimensional data analysis by providing an interpretation of feature relevance in visual clusters through interactive exploration of UMAP projections. Specifically, DimVis uses a contrastive EBM model that is trained in real time to differentiate between the data inside and outside a cluster of interest. Taking advantage of the inherent explainable nature of the EBM, we then use this model to interpret the cluster itself via single and pairwise feature comparisons in a ranking based on the EBM model's feature importance. The applicability and effectiveness of DimVis are demonstrated via a use case and a usage scenario with real-world data. We also discuss the limitations and potential directions for future research.

DimVis: Interpreting Visual Clusters in Dimensionality Reduction With Explainable Boosting Machine

TL;DR

DimVis tackles the interpretability challenge of nonlinear dimensionality reduction by employing a contrastive, supervised Explainable Boosting Machine (EBM) that is trained in real time on user-selected data to explain clusters in DR projections. The approach yields single- and pairwise-feature importances visualized through accessible plots, enabling rapid, interpretable insight into why clusters form. Demonstrations on healthcare datasets (e.g., Breast Cancer Wisconsin, Pima Indian diabetes) illustrate the method’s ability to reveal domain-relevant feature contributions and uncertainties, supporting more trustworthy DR-based exploration. This tool has practical implications for analysts seeking transparent, interactive explanations of complex high-dimensional data visualizations.

Abstract

Dimensionality Reduction (DR) techniques such as t-SNE and UMAP are popular for transforming complex datasets into simpler visual representations. However, while effective in uncovering general dataset patterns, these methods may introduce artifacts and suffer from interpretability issues. This paper presents DimVis, a visualization tool that employs supervised Explainable Boosting Machine (EBM) models (trained on user-selected data of interest) as an interpretation assistant for DR projections. Our tool facilitates high-dimensional data analysis by providing an interpretation of feature relevance in visual clusters through interactive exploration of UMAP projections. Specifically, DimVis uses a contrastive EBM model that is trained in real time to differentiate between the data inside and outside a cluster of interest. Taking advantage of the inherent explainable nature of the EBM, we then use this model to interpret the cluster itself via single and pairwise feature comparisons in a ranking based on the EBM model's feature importance. The applicability and effectiveness of DimVis are demonstrated via a use case and a usage scenario with real-world data. We also discuss the limitations and potential directions for future research.
Paper Structure (6 sections, 2 figures)

This paper contains 6 sections, 2 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. DimVis: System Overview and Use Case
  4. Usage Scenario: Interpreting Clusters in a Projection
  5. Discussion
  6. Conclusion

Figures (2)

  • Figure 1: Exploration of three clusters (C1--C3) produced by UMAP in (b) and local Feature Importances for the top three important features of the EBM model (c), trained on each cluster. (a) shows the dataset, UMAP hyperparameters, and EBM's precision and recall for C3. The most important feature (pair) per cluster is further explained in (d), with the selection mapped with a negative Score and the remaining points a positive one. For a single feature, Density histograms show the distribution of values. (e) reveals the GT labels used solely for verification.
  • Figure 2: The individual feature importance of SkinThick explains why the user-selected left and right subclusters in (a) and (b), respectively, were formed. The left subcluster (negative Score) contains points mostly with moderate values for SkinThick (Density) and low uncertainty (see the tight, gray band), while the right subcluster includes points with low values for SkinThick and relatively moderate uncertainty.