Sarcouncil Journal of Applied Sciences Aims & Scope

Abstract: The temporal variation of the maternal health indicators, such as blood pressure, fetal heart rate, glucose level, and gestational weight change, necessitates the use of predictive methodologies that can identify the temporal dependencies of the same. Static models can only capture sequential connections, irregular sampling, and multimodal signals and thus fail to provide adequate knowledge in early complications detection, e.g, preeclampsia, gestational diabetes, and preterm birth. Machine learning models that operate on sequences like recurrent neural networks, long short-term memory, gated recurrent units, and transformers solve these problems by enabling learning over complex patterns of time and the combination of heterogeneous forms of data. These models enable individual-based risk forecasting, active clinical actions, and population-wide analytics, whereas deployment via scalable, clear, and data protection-conscientious operations provides clinical reliability and conformity. This article summarizes the issues that affect temporal maternal data, the use of sequence-based models, the frameworks of deployment of these models, and the new available research directions such as multimodal integration, self-supervised pretraining, and cost-efficient, scalable approaches.