Unlocking the Power of Explainability in Ranking Systems: A Visual Analytics Approach with XAI Techniques

Ranking systems are widely used in various domains, including healthcare, to support decision making. However, understanding how these systems generate rankings can be challenging for users. In this paper, we present a visual analytic tool that combines XAI methods and interactive visualizations to explain ranking systems. Our tool provides users with a better understanding of how these systems work by using customized counterfactual explanations and feature importance visualizations. Unlike traditional counterfactual explanations that identify the minimum changes required to change class prediction, our tool considers a dynamic threshold that is set by other items in the list of items to be ranked. This threshold determines what changes are required for a specific item to rank lower or higher in the list. Our feature importance visualization shows the impact of each feature on the prediction, providing users with insights into how the system generates rankings. To demonstrate the effectiveness of our tool, we applied it to triage patients and rank them for admission to the ICU based on their severity. We demonstrated that our tool can provide clinicians with a better understanding of the ranking system and helped them make informed decisions about patient care. Our tool can also be applied to other ranking systems in healthcare and other domains, providing users with a transparent and understandable system for ranking-based decision support.