SVDD control charts based on MEWMA technique for monitoring compositional data

Journal: Computers & Industrial Engineering, WoS-SCIE, Q1
Author: Trần Kim Đức - IAD, Dong A University, 50000, Da Nang, Viet Nam

Abstract: Monitoring compositional data (CoDa) using control charts has become increasingly important in Statistical Process Control (SPC). This study introduces two approaches for CoDa monitoring, utilizing support vector data description (SVDD) control charts in conjunction with the multivariate exponentially weighted moving average (MEWMA) technique, specifically focusing on Phase II monitoring processes. The proposed approaches use two transformation methods: the Dirichlet density transformation and the isometric log-ratio transformation. We evaluate the effectiveness of the proposed SVDD control charts by computing the out-of-control zero-state Average Run Length (ARL1) using simulated data. Our results demonstrate that SVDD control charts detect anomalies more effectively than the traditional MEWMA control chart across various scenarios in monitoring CoDa. These findings contribute to the advancement of SPC and offer valuable insights for practitioners involved in CoDa monitoring across diverse applications.

DOI: https://doi.org/10.1016/j.cie.2025.110865