Electrical Impedance Spectroscopy Based Preterm Birth Prediction with Machine Learning

Preterm birth (PTB) is a critical global health concern as it stands as the leading cause of neonatal mortality. The objective of this research is to construct Machine Learning (ML) models functioning as a decision support system, intended to predict the likelihood of preterm delivery among pregnant women at high risk. We applied different rebalancing optimization methods to an extremely imbalanced electrical impedance spectroscopy (EIS) data set. Our EIS data set consisted of 365 records, and 12% ( n = 43) of records were preterm birth. We employed four classical machine learning classifiers, namely Logistic Regression, Random Forest, MultiLayer Perceptron, and Support Vector Machine, along with two distinct data rebalancing approaches: synthetic minority oversampling technique (SMOTE) and weighed balancing method. Our primary means of assessing model efficiency were sensitivity and AUC. Our findings indicated that SMOTE significantly enhanced the prediction performance of the PTB group, MLP attained the best sensitivity score of 0.89.