A1368
Title: Optimal kernel smoothing method for detecting fixed and random mean change in multivariate data
Authors: Yanhong Wu - California State University Stanislaus (United States) [presenting]
Abstract: To bolster the effectiveness of sequential detection methods, a kernel smoothing moving average (KSMA) chart is introduced that emphasizes recent observations. This novel approach is pitted against established methods like the finite moving average (FMA), the exponentially weighted moving average (EWMA), and the CUSUM charts, scrutinized through the lens of the conditional average delay detection time (ADDT) for a given average in-control run length (ARL0) in scenarios of persistent change, as well as the percentiles for transient change. A tailored kernel function aimed at optimizing efficiency within the spectrum of signal strength is proposed. The comparative analysis spans both fixed mean and random mean change models, encompassing both univariate and multivariate observations. Findings underscore that the proposed method not only outperforms the CUSUM and Shiryayev-Roberts (S-R) procedures at the specified signal strength reference value but also surpasses the performance of EWMA and FMA charts. Detection of both the fixed and random mean change in daily returns for Dow Jones Industrial Stocks is used for illustration.
