Pregnancy Risk Prediction Dataset from West Lombok, Indonesia

This dataset contains structured clinical data from 5,313 pregnant women collected from public health centers in West Lombok Regency, Indonesia, for pregnancy risk prediction research. The dataset addresses the critical need for automated early prediction of pregnancy complications to enable appropriate perinatal care and reduce maternal and neonatal mortality rates.The data was collected during a cohort study from February 2020 to February 2021, encompassing comprehensive maternal health indicators including demographic variables, anthropometric measurements, laboratory results, medical history, and clinical assessments. Each case was classified into four pregnancy risk categories based on standardized clinical protocols: normal pregnancy (65.7%), low-risk pregnancy (27.6%), moderate-risk pregnancy (5.6%), and high-risk pregnancy (1.1%).The dataset features 16 clinical variables: maternal age, parity, pregnancy spacing, height, hemoglobin levels, mid-upper arm circumference, hepatitis B surface antigen status, HIV status, diabetes mellitus history, hypertension history, proteinuria, blood pressure measurements, bleeding history, and comorbidity status. All data underwent rigorous quality assessment with clinical validation by experienced healthcare professionals following American Academy of Pediatrics and American College of Obstetrics and Gynecology guidelines.This dataset represents a valuable resource for developing and evaluating machine learning models for pregnancy risk stratification, particularly for addressing class imbalance challenges common in medical datasets. The severe imbalance between normal pregnancies and high-risk cases (58:1 ratio) makes this dataset particularly suitable for testing ensemble methods, cost-sensitive learning approaches, and specialized techniques for handling minority class detection in healthcare applications.Both the original raw dataset (5,324 records) and the cleaned dataset (5,313 records) are provided to support transparency and enable researchers to apply different preprocessing approaches. The small percentage of excluded records (0.2%) resulted from removing physiologically implausible values and records with missing essential clinical parameters, following established medical data preprocessing standards.This dataset has been successfully used to develop a novel ensemble XGBoost-DQN approach achieving 98.19% accuracy in pregnancy risk prediction, demonstrating its utility for advancing automated maternal health monitoring systems.