Machine Learning Based Adaptive EWMA Control Charts for Process Dispersion

Monitors the variance of a process with adaptive exponentially weighted moving average (AEWMA) control charts whose smoothing weight is predicted by a regression model fitted to the transformed subgroup variance. The subgroup variance is transformed as in Castagliola (2005) ; the smoothing weight is predicted by one of nine pre-fitted models from four families (neural networks, support vector machines, regression trees and boosting, and a quantile regression with lagged inputs), following the parameter-free approach of Noor-ul-Amin et al. (2024) , related work on support vector regression based adaptive EWMA charts by Kazmi and Noor-ul-Amin (2024) , and the adaptive EWMA idea of Capizzi and Masarotto (2003) . The comparison of the models is described in Atar, Purutçuoğlu and Bursalı (2026), accepted at the 3rd International Congress on Data Science and Information Technologies (INFTEC 2026). Control limits are calibrated by Monte Carlo simulation to a common in-control average run length of 370 for subgroups of size five. Functions are provided to run a chart on data, to compare all nine charts, and to estimate average run lengths by simulation.