Preservation of Feature Stability in Machine Learning under Data Uncertainty for Decision Support in Critical Domains

Aug 2, 2026

Upload by Skyler

Machine learning is increasingly used to support decisions in critical domains, where explainable, stable, and relevant inputs are essential. Reliable deployment requires understanding how missing data affects feature variability, especially when methods rely on an idealized feature set. The study closes this gap by running experiments with traditional machine learning against a recently deployed descriptive, human-mimicking classification method. Results show higher classification accuracy and improved feature-selection stability as data incompleteness increases, indicating descriptive approaches can strengthen uncertain decision support.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages30

Reading Time1 h 16 min

AccessFree to download

Introduction

Machine learning for critical decision-making

Missing data and its challenges

Explainable AI and feature stability

Proposed descriptive approach

Research motivation and scope

Methods

Algorithms used

Abstraction approaches and limitations

Experiments

Research problem and experimental design

Results

Findings on feature stability and accuracy

arXiv :2401 . 11044v2 [ cs .LG] 6 Aug 2024 Preservation of Feature Stability in Machine Learning under Data Uncertainty for Decision Support in Critical Domains Karol Capalaa,b,∗, Paulina Tworeka , Jose Sousaa a Personal Health Data Science Group, Sano - Centre for Computational Personalised Medicine, Czarnowiejska 36, 30-054 Krak´ow, Poland b Institute of Computer Science, AGH University of Krakow, Kawiory 21, 30-059 Krak´ow, Poland Abstract In a world where Machine Learning (ML) is increasingly deployed to support decision-making in critical domains, providing decision-makers with explainable, stable, and relevant inputs becomes fundamental. Understanding how machine learning works under missing data and how this affects feature variability is paramount. This is even more relevant as machine learning approaches focus on standardising decision-making approaches that rely on an idealised set of features. However, decision-making in human activities often relies on incomplete data, even in critical domains. This paper addresses this gap by conducting a set of experiments using traditional machine learning methods that look for optimal decisions in comparison to a recently deployed machine learning method focused on a classification that is more descriptive and mimics human decision making, allowing for the natural integration of explainability. We found that the ML descriptive approach maintains higher classification accuracy while ensuring the stability of feature selection as data incompleteness increases. This suggests that descriptive classification methods can be helpful in uncertain decision-making scenarios. Keywords: machine learning, critical scenarios, decision making, standardising, descriptive, uncertainty, missing data ∗ Corresponding author Email addresses: [k.capala@sanoscience.org](k.capala@sanoscience.org) (Karol Capala), Preprint submitted to Elsevier August 7, 2024 1. Introduction Advancements in Machine Learning (ML) are changing decision-making in critical domains, from healthcare to finance [1] . This increases the need to provide models capable of being integrated into the decision-making process. However, despite their potential, ML algorithms are often plagued by limitations, including lack of generality and explainability, bias, and outcome stability under missing data [2] . Furthermore, ensuring stable and explainable predictions under missing data is crucial for establishing the reliability required for ML to support decision-making in critical domains. Traditionally, ML addresses missing data using imputation or modelling techniques [3] . These approaches come with two challenges. Its implementation requires data preprocessing, which can be costly and particularly time-consuming and challenging when using low-quality datasets. Secondly, imputing missing values can introduce erroneous information into the model, especially with high missing value rates or biased incomplete patterns. Such challenges can cascade negative impacts on the entire model, weakening its ability to classify data correctly and potentially altering the identification of the most essential features. The adoption of ML methods to support decision-making in critical domains mainly focuses on increasing decision-making accuracy. However, it needs to provide an explainable rationale to be integrated into the decisionmaking process. Therefore, an ML methodology that maintains feature stability under missing data [4], while minimising the need for data preprocessing is crucial. Moreover, understanding the rationale behind a decision is fundamental for ML to support decision-making and forms the basis of explainable AI (XAI) [5 , 6 , 7] . To deal with the missing data without the need for imputation and data modelling, recently, an ML approach has been proposed by us [8 , 9] . The proposed approach demonstrated the ability to maintain a constant level of classification accuracy even with a very high

FAQs

What was the main experimental finding regarding accuracy and feature selection stability?

The descriptive classification approach maintained higher classification accuracy and ensured more stable feature selection as data incompleteness increased. This suggests the method is more robust in uncertain decision-making scenarios.

Aug 2, 2026

Upload by Skyler