Estimation of an Epidemic Curve During an Outbreak: A Classification Approach

• Activity Number: 480
• Type: Contributed
• Date/Time: Wednesday, August 4, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Elaine O. Nsoesie*+ and Richard Beckman and Madhav Marathe
• Companies: Virginia Tech and Virginia Tech and Virginia Tech
• Keywords: epidemic curves ; influenza ; classification ; support vector machines ; random forests ; nearest neighbor

Abstract

Classification techniques are widely used for identifying underlying groupings within datasets. We propose using classification as a means for estimating epidemic curves and inferring disease model parameters from surveillance data. Support vector machines, random forests and nearest neighbor are some of the classification techniques considered in our study. Our results suggest three main conclusions. First, classification techniques can be used to identify and make inferences about disease outbreaks based on partial epidemic curves. Second, classification techniques, which are easy to implement and consist of fewer assumptions regarding the structure of the epidemic curve, perform better than more complicated techniques. Finally, the epidemic curves for severe outbreaks are easier to estimate since the structures are usually distinct from other epidemic curves.