Algorithm Combination for Improved Performance in Biosurveillance
This chapter proposes an enhancement to currently used algorithms for monitoring daily counts of pre-diagnostic data. Rather than use a single algorithm or apply multiple algorithms simultaneously, our approach is based on ensembles of algorithms. The ensembles lead to better performance in terms of higher true alert rates for a given false alert rate. Combinations can be employed at the data preprocessing step and/or at the monitoring step. We discuss the advantages of such an approach and illustrate its usefulness using authentic modern biosurveillance data.
KeywordsControl charts Monitoring Ensemble methods Optimization Pre-diagnostic data
- Box, G., Luceno, A. (1997)Statistical Control: By Monitoring and Feedback Adjustment. 1st ed. Wiley-Interscience, London.Google Scholar
- Burkom, H. S. (2003) “Development, adaptation and assessment of alerting algorithms for biosurveillance”,Johns Hopkins APL Technical Digest24(4), 335–342.Google Scholar
- Lotze, T. and Shmueli, G. (2008) “Ensemble forecasting for disease outbreak detection”, 23rd AAAI Conference on Artificial Intelligence, Chicago July 08. Google Scholar
- Lotze, T., Shmueli G. and Yahav, I. (2007) Simulating Multivariate Syndromic Time Series and Outbreak Signatures, Robert H. Smith School Research Paper No. RHS-06-054, Available at SSRN: http://www.ssrn.com/abstract=990020.
- Lotze, T., Murphy, S. P., and Shmueli, G. (2008) “Preparing biosurveillance data for classic monitoring”, Advances in Disease Surveillance 6.Google Scholar
- Montgomery, D.C. (1997)Introduction to Statistical Quality Control.3rd ed. Wiley, London.NIST/SEMATECH: E-handbook of statistical methods.http://www.itl.nist.gov/div898/handbook.
- Shmueli, G. and Burkom, H. (2008) “Statistical challenges in modern biosurveillance”,Technometrics, in press. Google Scholar
- http://www.projectmimic.com: R code for simulating or mimicking multivariate time series of biosurveillance data and outbreak signatures. Also includes semi-authentic data.