Journal of Advanced Robotics, Autonomous Systems and Human-Machine Interaction
Predictive Modeling for Maternal Health Risk Assessment Using Wearable and IoT Data
Abstract
Idowu Olugbenga Adewumi
Maternal mortality continues to be a significant global health issue, with around 287,000 fatalities worldwide each year and about 34,000 deaths recorded in Nigeria in 2023. Despite improvements in antenatal care, more than 60% of maternal fatalities in sub-Saharan Africa result from preventable complications like preeclampsia (4 - 8% prevalence), gestational diabetes mellitus (5 - 10%), and preterm delivery (11 - 13%). This research offers a predictive modeling framework utilizing physiological data obtained from wearable’s and IoT, which encompasses heart rate variability (recorded at 1 Hz resolution), blood pressure (gathered every 15 minutes), oxygen saturation (SpO2, averaging between 92–99%), and glucose levels (collected every 6 hours) to identify early indications of maternal risk. We assess conventional machine learning classifiers (Logistic Regression, Random Forest) in comparison to sophisticated sequence models (LSTM, Bi-LSTM, Transformer encoder–decoder). Experiments performed on a synthetic but population-representative dataset comprising 49,969 longitudinal records from 2,000 Nigerian patients show that the Transformer model delivers enhanced performance with an accuracy of 94.2%, precision of 0.89, recall of 0.91, F1-score of 0.90, and ROC–AUC of 0.95, outperforming LSTM (90.8% accuracy, 0.91 ROC–AUC) and Random Forest (85.2% accuracy, 0.83 ROC–AUC). Temporal analysis shows that expanding sequence length from 10 to 50-time steps enhanced prediction accuracy by 3.7% for LSTM, 3.6% for Bi-LSTM, and 3.7% for Transformers, highlighting the significance of long-term observation. Evaluation of computational efficiency indicates training durations varying from 1.5 minutes (Logistic Regression) to 22.4 minutes (Transformer), while keeping inference latency below 2.1 ms for each sample, confirming practicality for real-time implementation. This work emphasizes the capability of IoT-enabled, ML-based maternal risk forecasting to lower mortality rates by as much as 30% in high-risk groups via early intervention strategies in healthcare settings, by integrating SDG 3 (Good Health and Well-being) and SDG 5 (Gender Equality) in both rural and urban areas.