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 Video

Updated: Jan 10, 2026

A Data-Driven Approach to Quantifying Immune States in Sepsis
07:42

A Data-Driven Approach to Quantifying Immune States in Sepsis

Published on: February 7, 2025

465

Improving Sepsis Mortality Prediction With Machine Learning Using Full Region Synthetic Sampling Approach.

Ibrahim A Amory1, Parviz Rashidi Khazaee2, Saleh Yousefi2

  • 1Computer Engineering Department Urmia University Urmia Iran.

Health Science Reports
|November 26, 2025
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 journal

Knowledge and Awareness of HPV, the HPV Vaccine, and Cervical Cancer Among Women From a Brazilian Community: A Community-Based Cross-Sectional Study.

Health science reports·2026
Same journal

Wearable Electronic Devices for Alzheimer's Disease Care: A Systematic Review of Applications, Benefits, and Challenges.

Health science reports·2026
Same journal

Time-of-Day Effects on Postural Balance in Blind Children: A Cross-Sectional Comparative Study.

Health science reports·2026
Same journal

Immediate or Early vs. Delayed Loading Protocols for Two-Implant Mandibular Overdentures in Edentulous Population: A Comprehensive Systematic Review and Meta-Analysis of Randomized Controlled Trials.

Health science reports·2026
Same journal

Determining the Organizational Culture and Readiness for Evidence-Based Practice Amongst Surgical Ward Nurses in Namibia: A Cross-Sectional Study.

Health science reports·2026
Same journal

BCR-ABL1 Transcript Variants in Chronic Myeloid Leukemia Patients in Africa: A Systematic Review.

Health science reports·2026

This study introduces novel data augmentation techniques, Full Region Synthetic Sampling Approach (FRSSA) and Adaptive Synthetic Sampling Tuning (ASST), to improve machine learning models for predicting sepsis mortality in Intensive Care Units (ICUs). The Post-Splitting augmentation strategy with FRSSA and ASST demonstrated superior fairness and reliability for clinical risk assessment.

Area of Science:

  • Medical Informatics
  • Machine Learning
  • Critical Care Medicine

Background:

  • Sepsis is a critical condition with high mortality rates in Intensive Care Units (ICUs).
  • Accurate prediction of sepsis mortality is vital for effective ICU resource management.
  • Class imbalance in ICU datasets significantly hinders the performance of machine learning models.

Purpose of the Study:

  • To develop and evaluate novel data augmentation methods for improving sepsis mortality prediction models.
  • To address the challenge of class imbalance in ICU datasets.
  • To propose a fair and clinically relevant evaluation metric for personalized risk assessment.

Main Methods:

  • Proposed the Full Region Synthetic Sampling Approach (FRSSA) for dynamic minority class balancing based on regional density.
Keywords:
Adaptive Synthetic Sampling Tuning (ASST)Balanced Performance Score (BPS)Full Region Synthetic Sampling Approach (FRSSA)MIMIC‐IVdata augmentationmachine learning in ICUsepsis mortality

More Related Videos

A Reproducible Intensive Care Unit-Oriented Endotoxin Model in Rats
05:56

A Reproducible Intensive Care Unit-Oriented Endotoxin Model in Rats

Published on: February 20, 2021

2.4K

Related Experiment Videos

Last Updated: Jan 10, 2026

A Data-Driven Approach to Quantifying Immune States in Sepsis
07:42

A Data-Driven Approach to Quantifying Immune States in Sepsis

Published on: February 7, 2025

465
A Reproducible Intensive Care Unit-Oriented Endotoxin Model in Rats
05:56

A Reproducible Intensive Care Unit-Oriented Endotoxin Model in Rats

Published on: February 20, 2021

2.4K
  • Introduced Adaptive Synthetic Sampling Tuning (ASST) to optimize augmentation weights for enhanced model performance.
  • Compared Pre-Splitting and Post-Splitting augmentation strategies, utilizing MIMIC-IV and eICU-CRD datasets with Random Forest, XGBoost, and LightGBM models.

    • Main Results:

    • The Pre-Splitting strategy showed high accuracy (89.64%) and AUROC (0.968) but risked test-set contamination.
    • The Post-Splitting strategy achieved 78.31% accuracy and 0.7241 AUROC, ensuring better real-world generalization.
    • ASST optimally balanced FRSSA interpolation (65%) and expansion (35%), reducing false positives and improving model fairness.

    Conclusions:

    • FRSSA preserves regional data distribution, while ASST dynamically adjusts augmentation for optimal performance and generalization.
    • The Balanced Performance Score (BPS) provides a flexible framework for evaluating models based on clinical priorities.
    • Post-Splitting augmentation using FRSSA and ASST yields a fairer and more reliable ICU mortality prediction model.