Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Fiper: a Visual-based Explanation Combining Rules and Feature Importance

Eleonora Cappuccio, Daniele Fadda, Rosa Lanzilotti, Salvatore Rinzivillo

TL;DR

The paper addresses the challenge of explaining AI decisions in high-stakes settings by proposing FIPER, a visual interface that merges rule-based explanations with feature-importance rankings to support human understanding. It formalizes rules and FI, and presents an interactive two-panel visualization where FI ranks features and rule predicates are shown with intuitive encodings. A user study comparing FIPER to text-based LORE outputs and an XAI Library visualization demonstrates that FIPER reduces error rates and is preferred for high-dimensional datasets, albeit with longer task times. The work advances explainable AI visualization by integrating rule predicates with FI and outlines future directions like selectable FI methods and counter-rule visualizations to broaden applicability.

Abstract

Artificial Intelligence algorithms have now become pervasive in multiple high-stakes domains. However, their internal logic can be obscure to humans. Explainable Artificial Intelligence aims to design tools and techniques to illustrate the predictions of the so-called black-box algorithms. The Human-Computer Interaction community has long stressed the need for a more user-centered approach to Explainable AI. This approach can benefit from research in user interface, user experience, and visual analytics. This paper proposes a visual-based method to illustrate rules paired with feature importance. A user study with 15 participants was conducted comparing our visual method with the original output of the algorithm and textual representation to test its effectiveness with users.

Fiper: a Visual-based Explanation Combining Rules and Feature Importance

TL;DR

The paper addresses the challenge of explaining AI decisions in high-stakes settings by proposing FIPER, a visual interface that merges rule-based explanations with feature-importance rankings to support human understanding. It formalizes rules and FI, and presents an interactive two-panel visualization where FI ranks features and rule predicates are shown with intuitive encodings. A user study comparing FIPER to text-based LORE outputs and an XAI Library visualization demonstrates that FIPER reduces error rates and is preferred for high-dimensional datasets, albeit with longer task times. The work advances explainable AI visualization by integrating rule predicates with FI and outlines future directions like selectable FI methods and counter-rule visualizations to broaden applicability.

Abstract

Artificial Intelligence algorithms have now become pervasive in multiple high-stakes domains. However, their internal logic can be obscure to humans. Explainable Artificial Intelligence aims to design tools and techniques to illustrate the predictions of the so-called black-box algorithms. The Human-Computer Interaction community has long stressed the need for a more user-centered approach to Explainable AI. This approach can benefit from research in user interface, user experience, and visual analytics. This paper proposes a visual-based method to illustrate rules paired with feature importance. A user study with 15 participants was conducted comparing our visual method with the original output of the algorithm and textual representation to test its effectiveness with users.
Paper Structure (16 sections, 4 figures)

This paper contains 16 sections, 4 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. Visual Explanation for Rules and Feature Importance
  4. XAI methods
  5. Visualization
  6. Other explanation modalities.
  7. User Study
  8. Participants and design
  9. The experimental tasks
  10. Procedure
  11. Data collection and analysis.
  12. Error rate.
  13. User satisfaction.
  14. Results
  15. Conclusions and future work
  16. ...and 1 more sections

Figures (4)

  • Figure 1: FIPER visualization of one instance of the German Credit Risk dataset. (Top) Attributes are sorted by the absolute value of FI. Categorical attributes are represented as stacked absolute bar charts. Numerical values are represented as box plots. The interval contained in the predicates of the rule are highlighted in yellow. (Bottom) Filtered view of the visualization, showing only the attributes referred in the rule premise
  • Figure 2: Finer details of a specific feature, selected by hovering the mouse on the corresponding row. (Top) Tooltip for a categorical data type, where the feature's actual value is shown with its class's cardinality. (Bottom) Tooltip for a numerical data type, where statistical central values are shown: min, max, median, Q1, and Q3.
  • Figure 3: The same instance of Figure\ref{['fig:fiper_example']} visualized as LORE output and XAI library visualization
  • Figure 4: Errors rate and completion times for tasks in each condition.(Top) The first column shows the absolute errors of the LORE output. The other two columns show the difference w.r.t. the first column, with divergent color scale to highlight increment or decrement in errors. (Bottom) Absolute number of errors for each output and each task