Optimized Feature Subset Selection Using Genetic Algorithm for Preterm Labor Prediction Based on...

Area of Science:

• Biomedical Engineering
• Obstetrics and Gynecology
• Machine Learning in Healthcare

Background:

• Preterm labor prediction remains a significant clinical challenge.
• Existing Electrohysterography (EHG) methods often rely on complex, difficult-to-interpret algorithms.
• There is a need for simpler, more interpretable EHG-based prediction models.

Purpose of the Study:

• To develop a preterm labor prediction system using Electrohysterography (EHG) with simple classification algorithms.
• To utilize genetic algorithms for optimizing feature selection from EHG data.
• To improve the interpretability and clinical applicability of EHG for preterm labor prediction.

Main Methods:

• Employed genetic algorithms to identify optimal feature subsets from 203 multichannel EHG parameters and obstetric data.
• Developed and validated three base classifiers: k-nearest neighbors, linear discriminant analysis, and logistic regression.
• Constructed an ensemble classifier combining base models to enhance prediction accuracy and reduce variability.

Main Results:

• Base classifiers achieved F1-scores ranging from 84.63% to 89.34%.
• Temporal, spectral, and non-linear EHG parameters across different bandwidths provided complementary predictive information.
• The ensemble classifier achieved a high F1-score of 92.04%, outperforming base models and demonstrating comparable performance to complex classifiers.

Conclusions:

• Simple, interpretable classification algorithms combined with optimized EHG features can effectively predict preterm labor.
• The developed ensemble model shows high generalization capacity, suitable for clinical implementation.
• This approach facilitates the translation of EHG technology into routine clinical practice for preterm labor prediction.