Early Prediction of Pregnancy Risk based on Cardiometabolic and Psychological Factors using Machine Learning

Early detection of high-risk pregnancies is critical for the improvement of maternal and fetal outcomes, especially in resource-constrained environments. In this study, we propose a machine learning-based method for classifying pregnancy risk levels (categorized as high or low) using a multidimensional dataset comprising cardiometabolic and psychological factors. A dataset with 1,205 observations was obtained from the Kaggle website, having features like Age, Systolic BP, Diastolic BP, Blood Sugar, Body Temperature, Previous Complications, Preexisting Diabetes, BMI, Gestational Diabetes, Heart Rate and Mental Health and Heart Rate. Data preprocessing was done using Synthetic Minority Oversampling Technique (SMOTE) for class balancing, one-hot encoding for handling categorical values and feature normalization. Twelve classification models were evaluated: Logistic Regression, K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Naïve Bayes, Decision Tree, Random Forest, XGBoost, AdaBoost, Gradient Boosting, CatBoost, LightGBM, and a multi-layer perceptron (MLP). Experiments were conducted using both 80:20 and 70:30 train–test splits. This study takes into consideration mental health data of the pregnant individual along with physical health data and finally introduces a predictive model based on Gradient Boosting giving 99.13% accuracy with 80:20 split for identifying high-risk pregnancies. The proposed solution offers a scalable, interpretable, and clinically relevant tool that can aid healthcare professionals in early risk detection and intervention planning, thus helping to improve maternal care and reduce mortality rates.