EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

Evaluating EMOO as a Metric-Balanced Ensemble Framework for Multi-Objective Optimization in Imbalanced Medical Data

Maryam Moradpour, Zully Ritter, Anne-Christin Hauschild

Studies in Health Technology and Informatics · 2026

DOI: 10.3233/shti261008 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Studies in Health Technology and Informatics, 2026. Verfasst von Maryam Moradpour und 2 weiteren Autor:innen. Beteiligte Einrichtung: Predictive Deep Learning in Medicine and Healthcare Department, Justus-Liebig University, Gießen, Germany. Hauptthemen laut MEDLINE: Medical Informatics. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Introduction

Class imbalance can hinder reliable detection of clinically relevant outcomes in binary clinical prediction. In this study, the positive class was predefined as the clinically relevant minority outcome; therefore, accuracy-only hyperparameter selection can favor the majority class and reduce sensitivity for that outcome.

Methods

We evaluated the previously published Ensemble Multi-Objective Optimization (EMOO) framework as a hyperparameter-selection and ensembling method rather than a diagnostic model itself. EMOO uses the Non-dominated Sorting Genetic Algorithm II (NSGA-II) to search Random Forest hyperparameter configurations across accuracy, sensitivity, specificity, and F1-score, while additionally minimizing their standard deviation (STD) to discourage imbalanced metric performance. Final Pareto-optimal models are combined through Adaptive Boosting (AdaBoost). EMOO was compared with baseline Random Forest, the Synthetic Minority Over-sampling Technique (SMOTE), and the Multi-Objective Optimization Framework (MOOF) on three public binary clinical datasets with imbalance ratios of 1.16–2.11, using repeated 10×5-fold stratified cross-validation and Decision Curve Analysis (DCA).

Results

EMOO achieved the highest sensitivity and geometric mean across all datasets and the highest specificity on the Heart Failure Clinical Records and Mammographic Mass datasets. On the Pima Indians Diabetes dataset, its higher sensitivity was accompanied by lower specificity than baseline Random Forest and MOOF. DCA showed higher net benefit than treat-all and treat-none strategies across relevant threshold ranges.

Conclusion

EMOO provides a metric-balanced approach to selecting and ensembling classifier configurations for moderately imbalanced clinical prediction tasks. External validation on independent clinical cohorts is required before routine use. The EMOO package and evaluation code are available at github.com/HauschildLab/EMOO and github.com/HauschildLab/EMOO4IMD.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Medical InformaticsHumansBoosting Machine Learning AlgorithmsClassification AlgorithmsGenetic AlgorithmsPrediction AlgorithmsRandom ForestSensitivity and Specificity

Stichwörter der Autor:innen

Boosting Machine Learning AlgorithmsGenetic AlgorithmsMachine LearningPrediction AlgorithmsSensitivity and Specificity

Autor:innen und Institutionen

  • Predictive Deep Learning in Medicine and Healthcare Department, Justus-Liebig University, Gießen, Germany.
  • Medical Informatics Department, University Medical Center Göttingen, Germany.
  • Emergency Department, University Medical Center Göttingen, Germany.

Zitieren

APA

Moradpour, M., Ritter, Z., & Hauschild, A. C. (2026). Evaluating EMOO as a Metric-Balanced Ensemble Framework for Multi-Objective Optimization in Imbalanced Medical Data. Studies in Health Technology and Informatics. https://doi.org/10.3233/shti261008

Vancouver

Moradpour M, Ritter Z, Hauschild AC. Evaluating EMOO as a Metric-Balanced Ensemble Framework for Multi-Objective Optimization in Imbalanced Medical Data. Studies in Health Technology and Informatics. 2026. doi:10.3233/shti261008

BibTeX
@article{Moradpour2026,
  author = {Moradpour, Maryam and Ritter, Zully and Hauschild, Anne-Christin},
  title = {Evaluating EMOO as a Metric-Balanced Ensemble Framework for Multi-Objective Optimization in Imbalanced Medical Data},
  journal = {Studies in Health Technology and Informatics},
  year = {2026},
  doi = {10.3233/shti261008}
}
Kostenlos starten →