Evaluating Directionally-Sensitive Multivariate Control Charts with an Application to Biosurveillance

摘要: The main goal of biosurveillance is the early detection disease outbreaks. Advances in technology have allowed collection, transfer, and storage pre-diagnostic information addition to traditional diagnostic data. Such data carry potential an earlier outbreak signature. In this work we deal with monitoring multivariate time series daily counts. Current temporal done univariately by applying control charts each separately. However, via has greatly reducing false alert rates increasing true rates. Classical are aimed at detecting shifts vector means any direction. Whereas one-side univariate easy obtain from their two-sided counterpart, directional sensitivity case non trivial. Several approaches were suggested for obtaining directionally-sensitive Shewhart chart (commonly referred as Hotelling T2 charts). there not been extensive comparison these methods it clear which approach performs better. compare two computational-feasible literature, namely Follmann's simple correction Testik Runger's (TR) quadratic programming approach. proposed directionally sensitive derive Multivariate EWMA (MEWMA) charts. We then perform analysis performance four methods, where examine model terms detection, robustness assumptions, training length characteristics. Our results show that TR's slightly better normally distributed data, yet more robust normality independence assumptions.