Introduction:
Maternal health complications continue to pose significant global challenges, particularly impacting maternal and infant morbidity and mortality in vulnerable populations. It is a fact that 287,000 maternal deaths were recorded in 2020, making maternal health a major public health issue worldwide, while, 94% of these deaths occurred in low- and middle-income countries. In recent years, advancements in artificial intelligence (AI) and machine learning (ML) have shown substantial potential in supporting early identification and management of such risks. AI plays an important role in obstetrics, as it contributes to early diagnosis, facilitating the prevention of complications and immediate intervention; it enhances the provision of personalized care, tailored to the unique needs, preferences, and lifestyle of each woman; and promotes the development of telehealth, offering access to personalized care for populations that often lack adequate health services.

Aims and Objectives:
Development and evaluation of machine learning models for the automatic classification of maternal risk (low, medium, high) through the use of physiological parameters (age, blood pressure, blood sugar, temperature, heart rate). To support midwives and healthcare professionals with timely and reliable risk assessment in order to improve maternal care.

Methods:
The study was based on a sample of 1014 pregnant women from Africa, selected to realistically reflect the needs and challenges of maternal care in the region. Data were collected from different healthcare sources, such as hospitals, community clinics, and maternal health centers, through an Internet of Things (IoT)-based risk monitoring system. This system enabled the continuous recording of vital parameters, the collection and processing of information in real time, and the early detection of potential complications.

Results:
Among the models evaluated, Random Forest demonstrated the most effective performance, achieving the highest scores in Accuracy (88.03%), True Positive Rate (88%), and Precision (88.10%).The application of the system showed high accuracy in identifying high-risk women (95.77%), confirming the effectiveness of artificial intelligence tools in obstetric care. The medium-risk category proved to be more challenging, as it is characterized by overlapping clinical features that make classification difficult and may require a combination of more indicators for reliable prediction.

Conclusions:
These results highlight its strength in classifying maternal health risks. However, all models struggled most with the mid-risk category, which exhibited lower Recall and Precision, emphasizing class imbalance as a key limitation in model performance. Early identification of women from all over the world at increased obstetric risk, before serious complications arise.