A Machine Learning Based Robust Adaptive EWMA Control Chart for Dispersion Monitoring

Statistical Process Control (SPC) is a popular method to check the stability of a process and identify abnormal changes. Nevertheless, the classical memory‐type graphs like the Exponentially Weighted Moving Average (EWMA) are very sensitive to outliers and data contamination, which may cause distortion of the monitoring statistics and decreases the reliability of its use. As a solution to this weakness, this paper will present a Machine Learning‐based robust Adaptive EWMA (ML‐RAEWMA) control chart to examine process dispersion in contaminated environments. The suggested plan incorporates anomaly‐based weighting, adaptive truncation, and dynamic smoothing in a strong EWMA model. A rolling median estimator gives protection against extreme values and a logistic weighting system based on anomaly scores restrains the impact of possible outliers. Comprehensive Monte Carlo experiments prove the statement that the ML‐RAEWMA chart maintains the nominal in‐control performance, but detects smaller and moderate dispersion changes more quickly than any current EWMA‐based scheme. Besides, the suggested chart exhibits a stable run‐length behavior with changing levels of contamination. The practical effectiveness of the method is further explained by a real‐data application. The findings verify that ML‐RAEWMA provides an effective and flexible monitoring model designed to be applied in contemporary processes that are associated with noise and data pollution.