Enhancing cardiotocography classification via ensemble learning and threshold optimization.

Area of Science:

• Medical Informatics
• Machine Learning
• Biomedical Engineering

Background:

• Healthcare datasets, particularly cardiotocography (CTG) data, often suffer from class imbalance.
• This imbalance leads to biased machine learning classifiers, resulting in poor performance on critical pathological cases.
• Existing research has overlooked optimizing classification thresholds as a solution for CTG data.

Purpose of the Study:

• To develop and evaluate a novel multifusion method for improving the classification accuracy of pathological cases in imbalanced CTG datasets.
• To address the limitations of current machine learning approaches in handling biased healthcare data.
• To enhance classification precision and maintain computational efficiency in fetal health monitoring.

Main Methods:

• A multifusion approach integrating undersampling techniques to balance the dataset.
• Incorporation of threshold-moving optimization to refine classification probability thresholds.
• Utilization of ensemble classifiers to aggregate predictions from multiple models.
• Application and validation on a dataset of 502 CTG cases from Czech Technical University and University Hospital Brno.

Main Results:

• The proposed multifusion method demonstrated significant improvements in identifying pathological cases compared to baseline models.
• Baseline models correctly classified approximately 2 out of 11 pathological cases per test.
• The enhanced approach achieved precision rates of 76.92%, 75%, and 41.67%, accurately identifying 9, 9, and 3 out of 12 pathological cases in respective tests.

Conclusions:

• The multifusion method effectively overcomes class imbalance and threshold issues in CTG data analysis.
• This approach offers a computationally efficient and precise solution for detecting pathological fetal conditions.
• The findings highlight the potential of integrating data balancing, threshold optimization, and ensemble methods for robust medical diagnostics.