A Hybrid Deep–Ensemble Intelligence Framework for Early Stratification of Pregnancy-Associated Risk Dynamics

Abstract

Maternal health complications are still a major concern worldwide, particularly in areas with limited medical resources where early risk identification is difficult. Existing methods mostly depend on manual checks and fixed threshold values, which often miss the complex and changing patterns seen during pregnancy. The system works with important health indicators such as age, blood pressure, glucose levels, and heart rate. To ensure that high-risk cases are properly identified, SMOTE is applied to balance the dataset. The model integrates a stacking approach, using RFC and GBC, along with an LSTM network to capture deeper patterns in the data. The final prediction is obtained by merging the outputs of both components through a soft voting method. The results show that this combined approach performs significantly better than individual models, achieving 99.03% accuracy along with equally strong precision, recall, and F1-score, while ETC and RFC achieve lower performance. The system is also designed with a simple Tkinter interface, making it easy for healthcare professionals to use it in real-time. This solution supports quicker and more accurate decision-making, helping doctors take early action and improve outcomes for both mothers and babies.