Abstrait

Identification of Concurrent Control Chart Patterns in Time Series

Shilpy Sharma, David A Swayne, Charlie Obimbo

Control chart patterns (CCPs) can be associated with certain assignable causes, and recognition of such patterns can assist the diagnostic search for those causes. Variations could be one or more instances of trend, cyclic, hugging, sudden shift or some other variations over time. Each pattern has special statistical characteristics which differentiate one pattern from another. In a time series, presence of more than one pattern may exist and identification of concurrent pattern is important. In this paper, we will utilize a new approach, RobustICA, for identification of concurrent patterns which is efficient when compared to traditional approaches being used for feature extraction. It will identify independent components hidden in mixture patterns and input those independent components to decision trees for recognition of as many as eight separate control chart patterns.