Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Optimizing stroke mortality prediction: a comparative study of sampling, cost-sensitive learning, and threshold

Rahim Nikbakht-Fard1, Leili Tapak2,3, Mojtaba Khazei4

  • 1Department of Biostatistics, School of Public Health, Hamadan University of Medical Sciences, Hamadan, Iran.

BMC Medical Informatics and Decision Making
|July 15, 2026
PubMed
Summary

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

A Comparative Analysis of the Immunoglobulin G and M Antibodies Seroprevalence Against Helicobacter pylori, Toxoplasma gondii, and Cytomegalovirus in Women With Preeclampsia.

American journal of reproductive immunology (New York, N.Y. : 1989)·2026
Same author

Impact of medical protective gown type and temperature conditions on cognitive performance of female healthcare workers during COVID-19.

Scientific reports·2026
Same author

Reduced expression of SOCS1-5 and STAT5a genes in Parkinson's disease patients: evidence for dysregulated immune signaling.

Molecular biology reports·2026
Same author

The Impact of Cognitive Behavioral Group Therapy with or without Lavender Aroma on Sleep Quality in Menopausal Women: Randomized Controlled Trial.

Current aging science·2026
Same author

Impact of advanced noise reduction algorithms on the diagnostic accuracy of vertical root fractures in cone-beam computed tomography: An evaluation of the influence of intracanal post types.

Imaging science in dentistry·2026
Same author

In Silico Transcriptomic Analysis for Identification of Potential Diagnostic and Prognostic Biomarkers and Therapeutic Targets in Cervical Cancer using a Hybrid Genetic Algorithm-Support Vector Machine Approach.

Archives of Iranian medicine·2026

Machine learning models, particularly XGBoost, show promise for predicting in-hospital mortality in stroke patients. Optimizing algorithms and thresholds significantly improves prediction accuracy, aiding early clinical decisions.

Area of Science:

  • Medical Informatics
  • Clinical Decision Support
  • Machine Learning in Healthcare

Background:

  • Stroke is a major global cause of death and disability.
  • Accurate prediction of in-hospital mortality is crucial for timely clinical interventions.
  • Class imbalance in mortality data presents a significant challenge for predictive modeling.

Purpose of the Study:

  • To compare various machine learning (ML) algorithms and imbalance-handling techniques for predicting in-hospital mortality in stroke patients.
  • To identify an optimal ML framework for stroke mortality prediction.
  • To evaluate the impact of different imbalance-handling strategies on model performance.

Main Methods:

  • Analysis of data from 2653 stroke patients.
  • Evaluation of five ML algorithms (XGBoost, RF, DNN, SVM, LR) using nested stratified 10-fold cross-validation.
Keywords:
Cost-sensitive learningImbalanced dataMachine learningSampling strategiesStroke

Related Experiment Videos

  • Application of threshold adjustment, cost-sensitive learning, and sampling methods (ENN, OSS, SMOTE, SMOTE-ENN, SVM-SMOTE) to address class imbalance.
  • Performance assessment using Accuracy, F1-score, G-mean, MCC, AUROC, and AUPRC.
  • Model interpretability analysis using SHAP.

    • Main Results:

    • XGBoost demonstrated the highest discrimination (AUROC 0.898 ± 0.014).
    • Threshold adjustment and cost-sensitive learning significantly improved G-mean (0.804 and 0.806, respectively) and F1-score.
    • Sampling methods like ENN and SMOTE-ENN showed favorable results for minority-class detection, though without significant statistical differences among them.
    • Key predictors identified include consciousness status, awareness level, arrival condition, Glasgow Coma Scale, respiratory complications, hospital stay duration, and age.

    Conclusions:

    • XGBoost is a highly effective algorithm for predicting in-hospital mortality in stroke patients.
    • Algorithm selection and threshold optimization are more critical than the specific sampling strategy for enhancing predictive performance.
    • Machine learning models, when optimized, offer significant potential for early risk stratification in stroke care.