A Semi-Supervised Machine Learning Approach in Predicting High-Risk Pregnancies in the Philippines

Area of Science:

• Maternal Health
• Machine Learning
• Public Health

Background:

• Early identification of high-risk pregnancies is vital for maternal and infant well-being.
• High fertility rates, particularly among youth in the Philippines, necessitate effective risk assessment tools.
• Existing supervised machine learning models face challenges with weak or scarce data.

Purpose of the Study:

• To compare supervised machine learning algorithms for predicting high-risk pregnancies using limited data.
• To address data scarcity challenges in high-risk pregnancy prediction.
• To develop and evaluate a semi-supervised approach for improved prediction accuracy.

Main Methods:

• Evaluated supervised learning algorithms (Decision Tree, Random Forest, SVM, KNN, Naïve Bayes, MLP) using 10-fold cross-validation and hyperparameter tuning.
• Applied a semi-supervised Self-Training model with a modified Decision Tree as the base estimator.
• Utilized a 30% unlabeled dataset alongside limited labeled data from Daraga, Albay, Philippines.

Main Results:

• The Decision Tree algorithm achieved the highest accuracy among supervised models, with a test score of 93.70%.
• The semi-supervised approach using the modified Decision Tree reached an accuracy rate of 97.01%.
• The semi-supervised model demonstrated superior performance compared to similar studies, especially in data-scarce scenarios.

Conclusions:

• Semi-supervised learning, particularly the Self-Training model with a Decision Tree base estimator, effectively improves high-risk pregnancy prediction accuracy with limited data.
• This approach offers a promising solution for regions with scarce data, enhancing maternal and infant care outcomes.
• The study highlights the potential of advanced machine learning techniques to address critical public health challenges in maternal health.