Nonparametric monitoring of high-dimensional processes via EWMA control charts based on random forest learning

• A novel nonparametric exponentially weighted moving average control scheme is developed to monitor high-dimensional data. • The random forest algorithm and log-likelihood method is used for dimensionality reduction. • The proposed control scheme outperforms its competitors in terms of various distributions and data types . • Two real case studies of the disk and breast cancer are adopted to investigate the effectiveness of our proposed chart. Multi-sensor systems have been widely used in manufacturing processes to perceive environmental or operating conditions and quality information, collecting large amounts of high-dimensional quality data. Because these systems are so common in practice, the important question has emerged as to how to leverage the full potential of rich data to monitor the various processes. Most traditional multivariate control charts in the field of high-dimensional data monitoring are no longer relevant because of their high dimensionality and typically unknown prior distribution of variables. In response to this need, this paper proposes a novel nonparametric Exponentially Weighted Moving Average (EWMA) control scheme by employing the random forest (RF) algorithm and log-likelihood method to transform high-dimensional data into one-dimensional data serving as the input of monitoring statistics. The simulation results indicate that the proposed control scheme outperforms its competitors in terms of various distributions and data types, especially for only one out-of-control (OC) cluster in the data stream. We also extend our scheme to the additional cases of a disk monitoring study and a breast cancer monitoring study prove the robustness and effectiveness of the proposed scheme.