Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

EXMOS: Explanatory Model Steering Through Multifaceted Explanations and Data Configurations

Aditya Bhattacharya, Simone Stumpf, Lucija Gosak, Gregor Stiglic, Katrien Verbert

TL;DR

This work introduces Explanatory Model Steering through ExMOS, a system that blends data-centric and model-centric global explanations to guide healthcare experts in configuring training data and improving diabetes-prediction models. Through quantitative and qualitative studies, the authors show that a hybrid explanation approach yields the best model improvement, albeit with higher perceived workload, while data-centric explanations better support understanding after configuration. The findings argue against relying solely on model-centric explanations and advocate for combined, layered explanations plus manual/data-driven configuration workflows. The study provides actionable design guidelines for explanation dashboards, data configuration mechanisms, and collaborative governance to support domain experts in high-stakes AI applications.

Abstract

Explanations in interactive machine-learning systems facilitate debugging and improving prediction models. However, the effectiveness of various global model-centric and data-centric explanations in aiding domain experts to detect and resolve potential data issues for model improvement remains unexplored. This research investigates the influence of data-centric and model-centric global explanations in systems that support healthcare experts in optimising models through automated and manual data configurations. We conducted quantitative (n=70) and qualitative (n=30) studies with healthcare experts to explore the impact of different explanations on trust, understandability and model improvement. Our results reveal the insufficiency of global model-centric explanations for guiding users during data configuration. Although data-centric explanations enhanced understanding of post-configuration system changes, a hybrid fusion of both explanation types demonstrated the highest effectiveness. Based on our study results, we also present design implications for effective explanation-driven interactive machine-learning systems.

EXMOS: Explanatory Model Steering Through Multifaceted Explanations and Data Configurations

TL;DR

This work introduces Explanatory Model Steering through ExMOS, a system that blends data-centric and model-centric global explanations to guide healthcare experts in configuring training data and improving diabetes-prediction models. Through quantitative and qualitative studies, the authors show that a hybrid explanation approach yields the best model improvement, albeit with higher perceived workload, while data-centric explanations better support understanding after configuration. The findings argue against relying solely on model-centric explanations and advocate for combined, layered explanations plus manual/data-driven configuration workflows. The study provides actionable design guidelines for explanation dashboards, data configuration mechanisms, and collaborative governance to support domain experts in high-stakes AI applications.

Abstract

Explanations in interactive machine-learning systems facilitate debugging and improving prediction models. However, the effectiveness of various global model-centric and data-centric explanations in aiding domain experts to detect and resolve potential data issues for model improvement remains unexplored. This research investigates the influence of data-centric and model-centric global explanations in systems that support healthcare experts in optimising models through automated and manual data configurations. We conducted quantitative (n=70) and qualitative (n=30) studies with healthcare experts to explore the impact of different explanations on trust, understandability and model improvement. Our results reveal the insufficiency of global model-centric explanations for guiding users during data configuration. Although data-centric explanations enhanced understanding of post-configuration system changes, a hybrid fusion of both explanation types demonstrated the highest effectiveness. Based on our study results, we also present design implications for effective explanation-driven interactive machine-learning systems.
Paper Structure (20 sections, 5 figures, 6 tables)

This paper contains 20 sections, 5 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Background and related work
  3. XAI Methods for ML Systems
  4. Model Steering Approaches in IML Systems
  5. Design of Explanations and Model Steering Approaches
  6. Explanation Dashboard
  7. Data Configuration Mechanisms
  8. Methods
  9. Study Design
  10. Prototype Implementation
  11. User Study 1: Quantitative Study
  12. User Study 2: Qualitative Study
  13. Results
  14. Discussion
  15. Combining Different Types of Explanations for Effective Data Configurations
  16. ...and 5 more sections

Figures (5)

  • Figure 1: Data-Centric and Model-Centric Explanation dashboards of our prototype
  • Figure 2: Hybrid explanation dashboard: combining data-centric and model-centric explanations from \ref{['fig:exmos_dce']} and \ref{['fig:exmos_mce']}
  • Figure 3: Data configuration screens from our prototype.
  • Figure 4: Box-plot showing the variation in the prediction accuracy scores obtained after the model steering task by participants of the three groups.
  • Figure 5: Box-plots showing variation of NASA-TLX scores between the three prototype versions and across the six evaluation areas of NASA-TLX such as mental demand, physical demand, time demand, performance, effort, frustration level.