Selecting Interpretability Techniques for Healthcare Machine Learning models - PDR框架下的八种算法综述

Aug 2, 2026

Upload by Asher

Healthcare decision-making benefits from interpretable machine learning that clarifies which data-derived or model-learned relationships drive predictions. Using the Predictive, Descriptive and Relevant (PDR) framework, the work distinguishes post-hoc methods that acquire interpretability after training from model-based methods where interpretability is embedded in the algorithm design. A curated overview presents eight candidate algorithms and maps them to key PDR characteristics, supporting clinicians in assessing recommendations with appropriate trust.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages26

Reading Time1 h 6 min

AccessFree to download

Introduction

Interpretability Definition and PDR Framework

Healthcare Relevance and High-Stakes Decisions

Two Approaches to Interpretability

Model-based (Intrinsic) Methods

Post-hoc Methods

Algorithm Selection and Classification

PDR Characteristics and Table Summary

Selecting Interpretability Techniques for Healthcare Machine Learning models Daniel Sierra-Botero1,2 , Ana Molina-Taborda1,2 , Mario S. Valdés-Tresanco1 , Alejandro Hernández-Arango3,4 , Leonardo Espinosa-Leal5 , Alexander Karpenko, Olga LopezAcevedo 1,2* 1. Biophysics of Tropical Diseases Max Planck Tandem Group, University of Antioquia UdeA, 050010 Medellin, Colombia 2. Grupo de Física Ató mica y Molecular, Facultad de Ciencias Exactas y Naturales, Universidad de Antioquia UdeA, 050010 Medellin, Colombia 3. Department of Internal Medicine, University of Antioquia, School of Medicine Medellin, Colombia 4. Hospital Alma Máter de Antioquia, University of Antioquia. Medellin, Colombia 5. Graduate School and Research, Arcada University of Applied Sciences, JanMagnus Janssonin aukio 1, 00560, Helsinki, Finland *Corresponding Author. Email: [olga.lopeza@udea.edu.co](olga.lopeza@udea.edu.co) Resumen: En el campo de la atención médica, se busca emplear algoritmos interpretables para ayudar a los profesionales de la salud en diversos escenarios de tomade decisiones. Siguiendo el marco Predictivo, Descriptivo y Relevante (PDR), se define el aprendizaje automático interpretable como un modelo de aprendizaje automático que de manera explícita y simple expone las relaciones determinadas en los datos oaprendidas por el modelo que son relevantes para su funcionamiento, y se categorizan los modelos como post hoc, adquiriendo interpretabilidad después del entrenamiento, obasados en el modelo, que están intrínsecamente integrados en el diseño del algoritmo. En este articulo presentamos una selección de ocho algoritmos, tanto post hoc como basados en el modelo, que pueden utilizarse para estos fines. Palabras claves: Aprendizaje de máquinas, Inteligencia Artificial, Interpretabilidad Abstract: In healthcare there is a pursuit for employing interpretable algorithms to assist healthcare professionals in several decision scenarios. Following the Predictive, Descriptive and Relevant (PDR) framework, the definition of interpretable machine learning as a machine-learning model that explicitly and in a simple frame determines relationships either contained in data or learned by the model that are relevant for its functioning and the categorization of models by post-hoc, acquiring interpretability after training, or model-based, being intrinsically embedded in the algorithm design. We overview a selection of eight algorithms, both post-hoc and model-based, that can be used for such purposes. Keywords: Machine Learning, Artificial Intelligence, Interpretability As the capabilities of machine learning models evolve, so does the urgency to understand the decision-making processes within them. This pursuit has led to an array of different related terms with various definitions depending on the authors and the context, such as explainability, transparency, intelligibility, etc. However, for the purpose of this article, we will adopt the definition of interpretability provided in the Predictive, Descriptive, and Relevant (PDR) framework. This framework defines interpretable machine learning as"the extraction of relevant knowledge from a machine-learning model concerning relationships either contained in data or learned by the model" (1) . This definition encloses many processes that can be categorized as interpretable and that vary drastically depending on the type of problem that is trying to be solved. Interpretable algorithms prioritize clarity and understanding over statistical inference. Instead of aiming for causal relationships, they emphasize the identification of correlations, which, while informative, can be less robust. In the context of healthcare decisions, as stated by Rudin and coworkers (2) , for high-stakes decisions, interpretability takes on a particular role. Rather than promoting blind trust in algorithmic outputs, it strives to establish a framework that empowers healthcare professionals, such as doctors, t

FAQs

What does the PDR framework define as interpretable machine learning?

It defines interpretable machine learning as extracting relevant knowledge from a model about relationships either contained in data or learned by the model that are relevant to its functioning.

How are post-hoc and model-based interpretability approaches different?

Post-hoc approaches apply after training to derive insights from the trained model, while model-based approaches integrate interpretability decisions into the training process so it is embedded in the algorithm design.

Why is interpretability especially important in healthcare?

For high-stakes decisions, interpretability helps avoid blind trust in algorithm outputs and enables healthcare professionals to evaluate recommendations and set an appropriate level of trust.

Aug 2, 2026

Upload by Asher