Distance-based outliers method for detecting disease outbreaks using social media

What is it about?

Forecasting the disease outbreaks could be useful for decision-making of public health resources. Social media provides a low-cost alternative source for public health surveillance. In this research we use Twitter data as a demonstration to detect influenza outbreak. We use distance-based outliers method to transform the noisy Twitter data into regions and then use regions to do region-based hypothesis testing for rapid outbreak detection. Majority voting has been used for decision making in committees. Our simulations show a good accuracy and robustness.

Featured Image

Why is it important?

In this research, we use Twitter data as a data source for public health surveillance. Twitter data can be noisy with nonstationary statistics. We bring a distance-based outliers method to disease outbreak detection. First, we compute outliers using distance-based method, and use the outlier identification as indicators that can break the data into regions which can be later subjected to hypothesis testing. A committee is then formed where each committee member has a preferred combination of the algorithm parameters (r; k;w): At the end, committee members vote and collectively decide on the best characterization of whether an outbreak has occurred and when. To our best knowledge this is the first work applying distance-based outliers method to convert temporal data into regions for detecting disease outbreak.