Research Project: Advancing Predictive Pathways for Maternal Mental Health Recovery: Machine Learning, Model Integration, and Scalable Deployment of Recovery Trajectories
Why did you apply for this internship?
I applied for this internship because it is a unique opportunity to work in a dynamic environment that combines data science, machine learning, and healthcare. What also excited me the most about this internship is the opportunity to contribute to research that supports an important cause.
What did you hope to gain in completing this project?
I hope to learn from a cross-disciplinary team and contribute to an impactful project as a data-driven problem solver. This opportunity will help me develop my technical and analytical skills while gaining valuable experience within a healthcare setting.
Project Overview
This project explores the development, integration, scalability, and deployment of machine learning models for decision-support in maternal mental health.
Using a synthetic dataset of over 100,000 records and 150+ features, EDA with spatial and socioeconomic analyses identified relationships between contributing factors and high-risk rural areas. Logistic Regression, Random Forest, SVM, and XGBoost models were optimised and combined into a soft voting ensemble to generate risk scores for maternal depression.
A maternal mental health dashboard, built with Flask and CSS, incorporates cloud-hosted models for perinatal safeguarding risk and worsening risk scores, presenting patient-level narratives alongside interactive visualisations. Together, these components create a scalable and interpretable framework that links technical innovation with patient-level recovery pathways in maternal mental health.
What were the key results of your research project?
- Processed a large-scale dataset, identifying how socioeconomic, biomarker, and psychosocial factors contribute to higher risk of maternal mental health concerns.
- Combined multiple model performances into a soft voting ensemble, achieving the most stable and robust performance, with a positivepredictive value (PPV) of 86% for maternal depression risk scores.
- Applied SHAP explainability to highlight feature importance, and ensure transparency in predictions for clinical interpretability.
- Developed a maternal mental health dashboard that integrates cloud-hosted machine learning models for worsening risk and rule-based perinatal safeguarding risk scores. It provides patient-level narratives, visualisations (e.g. time-series forecast), predictions, and integrates LLM (e.g. Gemini) to aid clinical review.
Overall, the project's results demonstrate that ensemble models can deliver robust predictions for maternal mental health risk scores. It also shows that combining ML models, safeguarding rules, and time-series forecasting within a scalable dashboard framework can generate interpretable, patient-level insights to support maternal mental health recovery pathways.
How do you feel you have benefited from completing this internship and has it made you consider future career paths?
Completing this internship has benefitted me in many ways.
It allowed me to consolidate my existing knowledge of data science and machine learning while also learning new concepts and applying them to a project with real-world impact. I also had the opportunity to strengthen my software engineering skills, which was challenging to do in a short timeframe, but an experience I thoroughly enjoyed.
The project opened my eyes to the kinds of problems faced in healthcare and how today's technological advances can support meaningful solutions. It has made me want to explore future career paths in data science, and beyond, like software engineering roles within healthcare, where further developing technological infrastructure can help support more people.
Download presentation slides: