Exploration of the Rashomon Set Assists Trustworthy Explanations for Medical Data

TL;DR

A novel method to explore models in the Rashomon set, extending the conventional modeling approach, and proposes the Rashomon_DETECT algorithm to detect models with different behavior, based on recent developments in the eXplainable Artificial Intelligence (XAI) field.

Abstract

The machine learning modeling process conventionally culminates in selecting a single model that maximizes a selected performance metric. However, this approach leads to abandoning a more profound analysis of slightly inferior models. Particularly in medical and healthcare studies, where the objective extends beyond predictions to valuable insight generation, relying solely on a single model can result in misleading or incomplete conclusions. This problem is particularly pertinent when dealing with a set of models known as $\textit{Rashomon set}$, with performance close to maximum one. Such a set can be numerous and may contain models describing the data in a different way, which calls for comprehensive analysis. This paper introduces a novel process to explore models in the Rashomon set, extending the conventional modeling approach. We propose the $\texttt{Rashomon_DETECT}$ algorithm to detect models with different behavior. It is based on recent developments in the eXplainable Artificial Intelligence (XAI) field. To quantify differences in variable effects among models, we introduce the Profile Disparity Index (PDI) based on measures from functional data analysis. To illustrate the effectiveness of our approach, we showcase its application in predicting survival among hemophagocytic lymphohistiocytosis (HLH) patients - a foundational case study. Additionally, we benchmark our approach on other medical data sets, demonstrating its versatility and utility in various contexts. If differently behaving models are detected in the Rashomon set, their combined analysis leads to more trustworthy conclusions, which is of vital importance for high-stakes applications such as medical applications.

Figures (6)

• Figure 1: Schematic overview of reasoning with Rashomon_DETECT algorithm. The ’caution’ field indicates potentially diverse models that ask for additional validation against domain knowledge. The ’explored’ field indicates the models with similar relationships. This method can be used to provide a trustworthy model for a given phenomenon.
• Figure 2: (Panel A) Distributions of dissimilarity metric values for the analyzed scenarios. Higher values reflect increased profile dissimilarity based on the corresponding measure. Note that the y-axes vary due to each measure's distinct theoretical value ranges. (Panel B) Example pairs of profiles generated for each examined scenario.
• Figure 3: The comparison of mean PDI measure for pairs: rf1 and each analyzed model. The darker the field is, the lower the measure is, meaning bigger differences between models. The following abbreviations refer to feature names: ALT (Alanine aminotransferase), ANC (Absolute Neutrophil Count), APTT (Activated Partial Thrombin Time), AST (Aspartate aminotransferase), Age, Bilirubin, CRP (C-Reactive Protein), Ferritin, Fibronectin, Fluid, Hb (Hemoglobin), PLT (Platelet Count), RBC (Red Blood Cell Count).
• Figure 4: Partial Dependence Plots for the detected variables with the highest PDI values.
• Figure 5: Partial Dependence Plots for continuous variables in medical data sets: (A) COVID data set, (B) PIMA data set. The lines represent PDPDs for various models from the Rashomon set. The three most distinct models detected for each data set are highlighted in red, green, and blue. The choice of metric in the Rashomon_DETECT process was based on profile shapes and scenarios analyses outlined in Section \ref{['measures_exp']}.

Theorems & Definitions (3)

• Definition 1: Reference model
• Definition 2: Rashomon set
• Definition 3: Profile