Predicting where COVID-19 spreads next with network analysis
- Article
- Health
More than a year after the first community quarantine in the Philippines, local authorities were still largely reacting to COVID-19 outbreaks after they appeared. A 2021 paper in Big Data Research by Carlos Andre Reis Pinheiro, Matthew Galati, Natalia Summerville, and Mark Lambrecht proposed something more proactive: use how people move between places to predict where cases will rise before they do.
The approach
The study used what the authors call location network analysis, or LNA, which pairs network analytics with machine learning. Two data sources feed it. Public health data gives the number of new positive cases per day per municipality. Telecommunications data gives the geolocation of anonymised mobile subscribers over time, which shows how the population moves between municipalities.
From the mobility data the authors built networks in which locations are nodes and movement between them forms the edges. They then applied standard network methods, including topological extraction, community detection, k-core decomposition, and centrality measures, to produce dynamic maps that identify the locations most important to the spread of the virus over time. A place that many people pass through, or that connects otherwise separate communities, is a place where an outbreak can seed many others.
The prediction model
The network features and the case history were used to train a gradient boosting model to predict which locations would have new cases the following week. The paper’s maps show the current week’s positive areas on one side and the predicted next-week areas on the other. On average the model’s accuracy was about 92 to 98 percent, with an overall sensitivity of around 90 percent. Accuracy counts true positives, true negatives, false positives, and false negatives together; sensitivity counts only how many of the real outbreak locations the model caught, which is the figure that matters most if the goal is to act before cases climb.
What authorities could do with it
The authors frame the results as a decision tool. Health officials could use the network measures to identify the level of risk in opening or closing particular locations, decide where shelter-in-place orders would do the most good, plan public transport service, and allocate medical resources toward areas expected to see a jump in cases. The post summarised it as an example of big data in healthcare that depends on combining fields: epidemiology, telecommunications, network science, and machine learning all had to be in the room.