No Single Metric Tells the Whole Story: A Multi-Dimensional Evaluation Framework for Uncertainty Attributions

TL;DR

This paper introduces a comprehensive multi-dimensional evaluation framework for uncertainty attributions in explainable AI, addressing inconsistencies in current assessment methods and providing a systematic way to compare different approaches.

Contribution

It aligns uncertainty attribution evaluation with the Co-12 framework, introduces new properties like conveyance, and demonstrates the framework's effectiveness through extensive experiments.

Findings

Gradient-based methods outperform perturbation-based in consistency and conveyance.

Monte-Carlo Dropconnect generally outperforms Monte-Carlo Dropout.

Most metrics show low inter-method agreement, indicating no single metric suffices.

Abstract

Research on explainable AI (XAI) has frequently focused on explaining model predictions. More recently, methods have been proposed to explain prediction uncertainty by attributing it to input features (uncertainty attributions). However, the evaluation of these methods remains inconsistent as studies rely on heterogeneous proxy tasks and metrics, hindering comparability. We address this by aligning uncertainty attributions with the well-established Co-12 framework for XAI evaluation. We propose concrete implementations for the correctness, consistency, continuity, and compactness properties. Additionally, we introduce conveyance, a property tailored to uncertainty attributions that evaluates whether controlled increases in epistemic uncertainty reliably propagate to feature-level attributions. We demonstrate our evaluation framework with eight metrics across combinations of uncertainty quantification and feature attribution methods on tabular and image data. Our experiments show that gradient-based methods consistently outperform perturbation-based approaches in consistency and conveyance, while Monte-Carlo dropconnect outperforms Monte-Carlo dropout in most metrics. Although most metrics rank the methods consistently across samples, inter-method agreement remains low. This suggests no single metric sufficiently evaluates uncertainty attribution quality. The proposed evaluation framework contributes to the body of knowledge by establishing a foundation for systematic comparison and development of uncertainty attribution methods.