Scholarlink

<p>Phase I analysis is the keystone of monitoring the process in Phase II. In case of incomplete Phase I data, the imputation method used in handling missing values plays a crucial role in the quality of the parameter estimates, which in turn leads to undermining the performance of control charts. Machine learning algorithms are praised for their ability to be trained without explicitly being programmed. They have outstanding performance in recognizing patterns in the data; hence they are used to address several problems, especially imputation. This study utilizes well-known machine learning algorithms: <i>k</i>-Nearest Neighbors (kNN), Support Vector Regression (SVR), and Random Forest (RF) in handling missing values in Phase I data and compares their performance to one of the state-of-the-art imputation methods, namely the Stochastic Regression (SRG) imputation. Our results show that the performance of the Phase I chart when machine learning imputation methods are used is far more powerful than that under the traditional ones in most cases. In other cases, they provide a similar performance to that of the traditional methods. The superiority of machine learning imputation methods is further illustrated by a practical application.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

On the Use of Machine Learning Imputation Methods in Phase I Multivariate Statistical Process Control

  • 期刊论文

  • 发表日期：

    2026年5月7日

查看全文
  • Dina A. Desoki,
  • Nesma A. Saleh,
  • Abdel Nasser Saad,
  • Mahmoud A. Mahmoud

摘要

Phase I analysis is the keystone of monitoring the process in Phase II. In case of incomplete Phase I data, the imputation method used in handling missing values plays a crucial role in the quality of the parameter estimates, which in turn leads to undermining the performance of control charts. Machine learning algorithms are praised for their ability to be trained without explicitly being programmed. They have outstanding performance in recognizing patterns in the data; hence they are used to address several problems, especially imputation. This study utilizes well-known machine learning algorithms: k-Nearest Neighbors (kNN), Support Vector Regression (SVR), and Random Forest (RF) in handling missing values in Phase I data and compares their performance to one of the state-of-the-art imputation methods, namely the Stochastic Regression (SRG) imputation. Our results show that the performance of the Phase I chart when machine learning imputation methods are used is far more powerful than that under the traditional ones in most cases. In other cases, they provide a similar performance to that of the traditional methods. The superiority of machine learning imputation methods is further illustrated by a practical application.