A0778
Title: Shapley values for identifying fault variables in MSPC
Authors: Sungim Lee - Dankook University (Korea, South) [presenting]
Juwhan Kim - Dankook University (Korea, South)
Johan Lim - Seoul National University (Korea, South)
Abstract: In multivariate statistical process control, Hotelling's T-squared (HT) control chart is widely used for detecting changes in the mean vector. However, its effectiveness is reduced if it fails to identify fault variables when an out-of-control status is signaled. A novel approach is presented using Shapley values, a powerful technique for explaining predictions in deep learning models. In general, Shapley values are usually estimated using the Shapley sampling method or the KernelSHAP algorithm. These methods are compared with existing methods, such as the Mason-Tracy-Young procedure and adaptive step-down procedure, which are prevalent in fault variable identification for HT control charts. The numerical studies demonstrate that the approach significantly outperforms existing methods in terms of sensitivity and specificity, particularly when changes are significant, but not remarkable. It is found that the approach improves fault variable identification.
