The Feature Understandability Scale for Human-Centred Explainable AI: Assessing Tabular Feature Importance

TL;DR

The paper tackles the challenge of aligning AI explanations with user understanding by introducing a validated Feature Understandability Scale for tabular data, with separate numerical and categorical versions that each manifest two factors. It follows a rigorous psychometric pipeline (Draft 1 through Draft 4) including expert interviews, focus groups, a pilot study, exploratory factor analysis, and confirmatory factor analysis, culminating in 8 numerical and 9 categorical items that form two coherent factors: Understanding and Measurement, and Feature-Outcome Relation. The scale demonstrates strong reliability and model fit, enabling per-feature understandability scoring and ranking to support an explainability-by-design workflow. This work provides a foundation for co-optimizing explainability and understandability in ML systems and suggests pathways for integrating understandability scores into feature selection and explanation generation in practice.

Abstract

As artificial intelligence becomes increasingly pervasive and powerful, the ability to audit AI-based systems is growing in importance. However, explainability for artificial intelligence systems is not a one-size-fits-all solution; different target audiences have varying requirements and expectations for explanations. While various approaches to explainability have been proposed, most explainable artificial intelligence methods for tabular data focus on explaining the outputs of supervised machine learning models using the input features. However, a user's ability to understand an explanation depends on their understanding of such features. Therefore, it is in the best interest of the system designer to try to pre-select understandable features for producing a global explanation of an ML model. Unfortunately, no measure currently exists to assess the degree to which a user understands a given input feature. This work introduces two psychometrically validated scales that quantitatively seek to assess users' understanding of tabular input features for supervised classification problems. Specifically, these scales, one for numerical and one for categorical data, each with two factors and comprising 8 and 9 items, aim to assign a score to each input feature, effectively producing a rank, and allowing for the quantification of feature prioritisation. A confirmatory factor analysis demonstrates a strong relationship between such items and a good fit of the two-factor structure for each scale. This research presents a novel method for assessing understanding and outlines potential applications in the domain of explainable artificial intelligence.

Figures (5)

• Figure 1: Overview of the xAI workflow for tabular data. The current work contributes to Step 1 by providing a way to measure 'understandability' for each feature.
• Figure 2: Overview of stakeholder taxonomies.
• Figure 3: Workflow of the Psychometric Development of the Scale
• Figure 4: Result of the CFA for the a) Numerical Scale b) Categorical Scale. The Abbreviations are as follows: Otp (Feature-Outcome Relation) and MnU (Understanding and Measurement).
• Figure 5: The assignment of the items to factors after the initial EFA for the a) Numerical Scale and b) Categorical Scale.