Data Science in Epidemiology

Spatio-temporal modelling and counterfactual analysis of the spread of COVID-19 using Mobility and Exposure

Mission of the project

Modeling localized disease patterns using multiple data sources is a challenging problem for epidemiologists. Further, capturing spatial relations and temporal dynamics using individual-level mobility data is also an important open problem. We propose to tackle this problem by combining SafeGraph dataset for mobility and exposure as well as the New York City Health department's report of weekly cases at the level of zip code to model zip code level COVID-19 infection spread. Finally, we aim to identify local hotspots for infections by directly accounting exposure of populations to POIs in our network-based metapopulation population.

Aggregation of mobility and infection datasets

We make use of widely used public datasets such as moblity network data from SafeGraph and infection data from NYCHealth.

Generating metapopulation graph for disease spread

We use this data to create a Bipartite matching from Regions (Zip Code Tract) to Places of Interest (POIs).

Neural Net Framework to predict Case-rates

We then create a neural framework that leverages past infection sequences in communities and movement information to POIs to predict future infections as well as contributions from POIs.

Hotspot detection

Using predictions from the model, we infer POIs that contribute most to infections to all regions. Identifying such hotspots could reduce strain on workers and efficiently direct budget to recovery.

Find a copy of our work here!

GitHub

EPI8803_Project.pdf

Page updated

Google Sites

Report abuse