Storytelling with Data for Improving Population Health

Retrieval-based machine learning (ML) models enable data to be supplemented by relevant information, retrieved from auxiliary databases and "memories". This enables the explainability of decisions made by the model and makes previously acquired knowledge accessible to human decision-makers.

In healthcare, routinely collected data typically cannot be entirely relied upon to make predictions for specific adverse events. Furthermore, explainable aspects (such as feature attribution) can be misinformed by the complex and incomplete nature of the data. This project will aim to:

i. Develop a ML model that uses retrieval-based query augmentation to build outcome predictions based on a multitude of linked data sources. ii. Develop a visual analytics interface that supports feature engineering and decision provenance.