Prediction of variables that the cause of out of control condition on multivariate control chart by ensemble machine learning algorithm

Multivariate control charts enable assessment of multi variable on a single chart rather than evaluating them individually. This control chart has the great advantage of not only saving time and workload, but also evaluating the relationships between variables. Contrary to these advantages, there is a disadvantage of not being able to determine which variables arise out of control. However, it should be known which corrective actions should be applied to the variable(s) in order to control the process. Supporting scientific methods are needed in this regard. Statistical and machine learning techniques are available in the literature. Machine learning methods have been used because of the lack of statistical methods to predict future situations. In this study, a machine learning based model has been developed to prediction of variables that the cause of out of control condition to eliminate this problem. The classification accuracy of the model is aimed to be as high as possible. In order to increase the accuracy of predictions, the basic single machine learning algorithms produce solutions with the most optimum parameters, and ensemble machine learning algorithms aiming to increase the accuracy by combining the algorithms have been used. The decision tree bagging and boosting methods, which were found to be the most successful among the 5 basic single algorithms, were combined separately and the accuracy increased. In the developed model, these two algorithms were combined with the stacking method and the other two machine learning algorithms were used together. Such use of nested ensemble algorithms is thought to improve the prediction accuracy. In order to prove the success of the model, it was applied in real life. The proposed model has been compared with the single machine learning algorithm, and two ensemble algorithms to prove the success of the study. By the help of the developed model, benefits such as consideration of time, cost and the relationships between variables have been obtained by use of multivariate control chart. In addition to fast diagnosis of the cause of out of control condition from new samples with high accuracy up to 98.06%.