Synthesizing Explainability Across Multiple ML Models for Structured Data

Area of Science:

• Machine Learning
• Explainable AI (XAI)
• Data Science

Background:

• High-stakes domains require reproducible feature importance aggregation across multiple models.
• Existing methods struggle to capture complex relationships in explainer outputs.

Purpose of the Study:

• To introduce the Weighted Importance Score and Frequency Count (WISFC) framework for robust ensemble feature-importance ranking.
• To provide a principled approach for reconciling and aggregating feature importance from diverse explainers.

Main Methods:

• The WISFC framework aggregates ranked outputs from diverse explainers.
• It assigns a weighted score based on rank and frequency across model-explainer pairs.
• This method consolidates weak signals from multiple modeling runs.

Main Results:

• WISFC generates a robust ensemble feature-importance ranking.
• It highlights consistently important features by aggregating diverse model perspectives.
• The framework offers a more principled approach than simple consensus methods.

Conclusions:

• WISFC enhances the exploration of complex systems by systematically combining multiple modeling perspectives.
• The framework is reproducible and generalizable for various machine learning models.
• It offers a novel strategy for researchers and practitioners in feature importance analysis.