Pregnancy Risk Prediction and Fetal Health Monitoring using Hybrid Machine Learning

Abstract

Monitoring pregnancy risks and fetal health is complex and requires careful attention, as newlinecomplications can be life-threatening for both the mother and baby. Obstetricians traditionally use manual methods, like tracking Fetal Heart Rate (FHR) with cardiotocography (CTG) and newlineanalyzing scans to assess risks. However, manual processes are prone to errors, which has led to growing interest in automated systems for more accurate and timely monitoring. newlineIn the first study, The author considered machine learning(ML) for risk prediction of newlinemother health and pregnancy formation by detecting ectopic pregnancy in the first trimester using XGBoost, utilizing data of 2,582 patients. The proposed model has the capability of handling bias-variance in dataset, handling missing values automatically, regularize parameters newlineavoiding over fitting and reduced time execution with parallel computing. The proposed model outperformed different Machine Learning algorithms namely Logistic Regression (LR), Naive bayes(NB), Random Forest(RF), Decision Tree (DT) and Support Vector Machine (SVM). newlineThe author, in the second work proposed an ML model to predict the risk in fetal growth newlineduring first and second trimester using FHR. The proposed LightGBM model used Grid newlinesearch-based tuning of hyper parameters on the FHR data for forecasting fetal growth as newlinenormal or abnormal using FHR. LightGBM avoids over-fitting by trimming trees using grid newlinesearch at the maximum tree depth and employing multi-threaded optimization to boost productivity and save time. The model used CTG dataset of 2,216 patients from Kaggle. newlineLater, the author proposed a stack ensemble model combining the predictions of multiple newlinebase models, resulting in better predictive performance than any single model alone. Hybrid models, namely DT, RF, Boosting, AdaBoost, XGBoost, and LightGBM, were considered as base models, with LR used as a meta-learner to classify fetal conditions as normal or abnormal. newlineThe author explored 12 ways of combining the base learners and found that the stack mode