Figure 1. Downtown Manhattan, New York City, where high population density makes infectious disease spread easily.

Predicting Flu With Mobility Behaviors

Gene Yang ‘19

Figure 1. Downtown Manhattan, New York City, where high population density makes infectious disease spread easily.

In the mathematical modeling of infectious disease, determining the mobility of infectious diseases as well as the mobility patterns of individuals in a population is crucial to predicting the spatial and temporal diffusion of such illnesses. Researchers from the University of Trento gathered two types of data, mobility data and self-reported health data, to construct and validate a model that predicts the future presence of such symptoms in individuals.

The first type of data, mobility data, was collected from 29 individuals via a mobile phone software that tracked communication events (e.g. calls and SMSs) and the participant’s location using GPS. Using this data, mobility features, including total distance traveled, maximum displacement between two visited places, and the radius of gyration, were then computed for each individual. The second type of data was a daily eight-question questionnaire completed by the same individuals over the span of one month. The questionnaire consisted of self-reported health symptoms with yes or no responses (e.g. shortness of breath, headache, muscle pain).

The model was then constructed on the basis of binary classification, a statistical concept that groups new observations, or individuals, into categories—whether those individuals reported yes or no for each question. For this classification task, three different machine learning models were used and compared: logistic regression (LR), random forest (RF), and gradient boosted trees (GBT). Results portrayed that the GBT model had the most predictive accuracy, with a AUCROC value of 0.62, which refers to the probability of the model to correctly classify individuals with or without flu-like symptoms. This study is one of the first that utilizes inference algorithms to predict health solely on the basis of mobility behaviors. However, as a result of the limited number of participants, 29 individuals, and the short duration of four weeks, more likeminded studies will be needed to accurately predict the relationships between mobility and flu-like symptoms.


  1. G. Barlacchi, et al., Are you getting sick? Predicting influenza-like symptoms using human mobility behaviors. DPJ Data Science 6, 6-27 (2017). doi: 10.1140/epjds/s13688-017-0124-6
  2. Image retrieved from:

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s