Center for Automated Learning and Discovery
School of Computer Science, Carnegie Mellon University
Framework for Using Grocery Data for
Early detection of epidemics and bio-terrorism attacks is of great concern to public health. There are many sources of data that might beused for early detection. The main goal of this project is to investigate the possibility of detecting epidemics and bio-terrorism early by analyzing trends in consumer grocery purchases. This type of data has two main advantages: first, we expect grocery data to have an earlier signal of an outbreak, since people tend to seek self treatment of symptoms before they reach a doctor or a hospital. Second, grocery data are much richer and are available on a more re ned scale than epidemiological data, that were previously the main source of information used for detection purposes. This paper introduces a framework that combines techniques from signal processing, forecasting and quality control to increase the sensitivity of the system and the rate of detection. Over-the-counter medication purchases were extracted from grocery data and used for experiments. This paper shows that grocery data can be used for the timely detection of epidemics and bio-terrorism attacks. The automated framework developed here is a first step towards a new generation of systems based on non-speci c syndrome data, i.e. data collected for purposes other than medical analysis. There is a hope that this work will have signi cant impact on the evolution of early detection and consequently on computer-based surveillance (CBS) systems.