Data-Driven Prediction of Maternal Health Risk Level Using Machine Learning Algorithms

Identification of maternal health risks is crucial due to the direct impact on the fetus. Pregnancy-related complications can be reduced through quality care and continuous health monitoring for expectant mothers. However, identifying risk levels and providing appropriate treatment remains challenging, particularly in resource-limited settings. A major contributor to maternal and child morbidity and mortality is the failure to identify risk levels at an early stage. To address this issue, we conducted extensive statistical data analysis and applied machine learning techniques to determine maternal risk levels (High or Low). The study utilized data from 1,205 pregnant women collected from Kurigram General Hospital, Bangladesh. Exploratory data analysis (EDA) was performed to uncover hidden patterns in the dataset. Missing values in continuous features were imputed using a skewness- and kurtosis-based method—KDE-MICE—while logistic regression was used to impute missing values in categorical features. Outliers were detected using two methods: Z-score and Inter-quartile Range (IQR), and were treated by replacing them with median values. Following data preprocessing, three feature selection techniques—Mutual Information (MI), correlation analysis, and permutation feature importance (PFI)—were employed to identify key risk factors. Significant predictors for maternal risk level included GD, pre-existing diabetes, BMI, HR, and mental health. Eight ML algorithms—LR, DT, RF, SVM, KNN, NB, GB, and AdaBoost were used for maternal risk prediction. GB based model demonstrated the highest performance: 98.17% accuracy, 98.10% precision, 98.09% recall, and a 98.09% F1-score.