Evaluating Explanatory Capabilities of Machine Learning Models in Medical Diagnostics - A Human-in-the-Loop Approach

Aug 3, 2026

Upload by Aria

This paper presents a comprehensive evaluation of machine learning explainability for medical diagnostics, focusing on Decision Trees, Random Forest, and XGBoost using a pancreatic cancer dataset. Human-in-the-loop methods and medical guidelines are leveraged to identify and prioritize clinically relevant features, serving both as a dimensionality reduction mechanism and as a basis for explainability assessment. Both agnostic and non-agnostic techniques are considered, and similarity measures such as the Weighted Jaccard Similarity coefficient are proposed to support interpretation. The aim is to select models that both perform well and align with human domain knowledge.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages35

Reading Time1 h 28 min

AccessFree to download

Abstract

Keywords

Introduction

Explainable AI (XAI) and benefits

Transparent vs opaque models

Need for validation in medical settings

arXiv :2403 . 19820v1 [ cs .LG] 28 Mar 2024 Evaluating Explanatory Capabilities of Machine Learning Models in Medical Diagnostics: A Human-in-the-Loop Approach Jos´e Bobes-Bascar´an 1 , Eduardo Mosqueira-Rey 1*,´Angel Fern´andez-Leal 1 , Elena Hern´andez-Pereira 1 , David Alonso-R´ıos 1 , Vicente Moret-Bonillo 1 , Israel Figueirido-Arnoso 1 , Yolanda Vidal- ´Insua2 1* Department of Computer Science and Information Technologies, University of Coru˜na (CITIC), R´ua Maestranza, 9, La Coru˜na, 15001, Galicia, Spain. 2* Servicio de Oncolog´ıa M´edica, Complejo Hospitalario (CHUS), R´ua da Choupana, s/n, Santiago de Compostela, 15706, Spain. *Corresponding author(s) . E-mail(s): [eduardo@udc.es](eduardo@udc.es) ; Contributing authors: jose.bobes@udc.es; [angel.fleal@udc.es](angel.fleal@udc.es) ; [elena.hernandez@udc.es](elena.hernandez@udc.es) ; [david.alonso@udc.es](david.alonso@udc.es) ; [vicente.moret@udc.es](vicente.moret@udc.es) ; [israel.figueirido.arnoso@udc.es](israel.figueirido.arnoso@udc.es) ; [yvidalinsua@gmail.com](yvidalinsua@gmail.com) ; Abstract This paper presents a comprehensive study on the evaluation of explanatory capabilities of machine learning models, with a focus on Decision Trees, Random Forest and XGBoost models using a pancreatic cancer dataset. We use Humanin-the-Loop related techniques and medical guidelines as a source of domain knowledge to establish the importance of the different features that are relevant to establish a pancreatic cancer treatment. These features are not only used asa dimensionality reduction approach for the machine learning models, but also as way to evaluate the explainability capabilities of the different models using agnostic and non-agnostic explainability techniques. To facilitate interpretation of explanatory results, we propose the use of similarity measures such as the Weighted Jaccard Similarity coefficient. The goal is to not only select the best performing model but also the one that can best explain its conclusions and aligns with human domain knowledge. 1 Keywords: Explainability, XAI, Machine Learning, Jaccard Similarity, Pancreatic Cancer 1 Introduction Explainable AI (XAI) [1] is a research field focused on making Artificial Intelligence (AI) systems in general, and Machine Learning (ML) systems in particular, more understandable to humans. Explainable AI offers several advantages, to name a few: it fosters confidence in the prediction of the model by making the decision-making process more transparent, promotes responsible AI development, aids in debugging and identifying issues, and allows auditing of AI models and checking if they adhere to regulatory standards. The inherent explainability of AI systems has not remained static but has changed considerably as a result of technological progress. In fact, explainability has become an increasingly difficult issue to tackle, as the internal functioning of AI systems has become less intelligible as they have become more complex [2] . Initially, symbolic AI models were explainable per se, e.g., rule-based expert systems could easily show to their users which rules they had followed to make a given decision, even though the rules can incorporate measures of uncertainty and imprecision as, for example, in fuzzy systems. These type of AI models are considered transparent, which means that the model itself is understandable [3], being understandability the characteristic of a model to make a human understand its function without any need for explaining its internal structure or the algorithmic means by which the model processes data internally [4] . However, rule-based systems presented too many limitations that hindered their development: they were rigid systems, with poor scalability and limited learning capabilities. To overcome the problems associated with symbolic AI, machine learning models were developed. These models can learn from data and improve their performance over time wit

FAQs

Which machine learning models are evaluated in this study?

The study evaluates Decision Trees, Random Forest, and XGBoost models on a pancreatic cancer dataset.

How does the paper incorporate human knowledge into the evaluation?

It uses Human-in-the-Loop related techniques together with medical guidelines to establish the importance of different features relevant to pancreatic cancer treatment.

What measure is proposed to help interpret explanatory results?

The paper proposes similarity measures such as the Weighted Jaccard Similarity coefficient to facilitate interpretation of explanatory outputs.

Aug 3, 2026

Upload by Aria