Safe Mom | Devpost

Safe Mom: Predictive Machine Learning for Preeclampsia

Inspiration

Maternal mortality remains one of the most pressing public health challenges globally, and Kenya is among the top 10 countries contributing to neonatal deaths. One key contributor to these deaths is preeclampsia, a life-threatening hypertensive disorder during pregnancy. We were particularly moved by the statistic that 20% of maternal deaths in Kenya are due to this preventable condition.

As a team of technologists and researchers passionate about using data for social good, we saw an opportunity to leverage machine learning to tackle this critical issue. Our vision was simple: what if we could predict the risk of preeclampsia early enough for healthcare providers to intervene? That single question sparked the birth of Safe Mom.

What it does

Our innovation, Safe Mom, leverages machine learning to predict the likelihood of preeclampsia early in a pregnancy, enabling healthcare providers to intervene sooner. With our AI model, clinicians can take proactive steps, reducing the risk of complications.

How we built it

We created a web-based platform that enables clinicians to assess the risk of preeclampsia using patient vitals. Here’s a breakdown of the tech stack:

Machine Learning: We used the XGBoost algorithm due to its high performance with structured data.

Backend: Built with Flask and FastAPI to serve predictions via a RESTful API.

Frontend: HTML/CSS was used to design a simple and accessible user interface.

Database: A MySQL database stores login credentials and patient records securely.

Integration: Our ML model is served via API, enabling real-time predictions during antenatal visits. All the code is available on https://github.com/SebbieMzingKe/Safe-Mom

Challenges we ran into

Data Scarcity Accessing clean and comprehensive healthcare datasets was difficult. We had to simulate some datasets for model training and testing while ensuring that our design would generalize well when real-world data is integrated.

Balancing Accuracy and Interpretability Ensuring our model was not just accurate, but also explainable, was essential since healthcare professionals rely on transparency for clinical decisions.

Clinician Adoption Through interviews, we realized that any tool we built needed to integrate seamlessly into existing clinical workflows. Simplicity and usability were as critical as accuracy.