We develop a mixture procedure to monitor parallel streams of data for achange-point that affects only a subset of them, without assuming a spatialstructure relating the data streams to one another. Observations are assumedinitially to be independent standard normal random variables. After achange-point the observations in a subset of the streams of data have nonzeromean values. The subset and the post-change means are unknown. The procedure westudy uses stream specific generalized likelihood ratio statistics, which arecombined to form an overall detection statistic in a mixture model thathypothesizes an assumed fraction $p_0$ of affected data streams. An analyticexpression is obtained for the average run length (ARL) when there is no changeand is shown by simulations to be very accurate. Similarly, an approximationfor the expected detection delay (EDD) after a change-point is also obtained.Numerical examples are given to compare the suggested procedure to otherprocedures for unstructured problems and in one case where the problem isassumed to have a well-defined geometric structure. Finally we discusssensitivity of the procedure to the assumed value of $p_0$ and suggest ageneralization.
展开▼