Grid-based K-nearest Neighbor Approach for Process Monitoring with Large Size Data

Abstract

Society today is experiencing an ever-increasing growth of the service industry. However, the use of statistical process control(SPC) within the service industry is not common. The application of SPC to the service industry can be just as beneficial as it is to the manufacturing industry, in improving service quality and ultimately customer satisfaction. The most common process control technique has been control charting. An anomaly score-based control chart is an algorithm that combines data mining and control charts to find the difference between normal values and outliers. Recently, k-nearest neighbor (kNN)-based outlier score calculation algorithms have been extensively studied, but due to the computational complexity of the kNN algorithm, its application to large amounts of data is limited. The purpose of this study is to propose a control chart using a grid-based kNN algorithm to reduce the time for the kNN algorithm to determine k-nearest neighbors. In order to verify the effectiveness of the proposed methodology, we experimented with changing hyper-parameters in various environments. Experimental results confirm that the methodology presented in this study reduces computation time while ensuring predictable and acceptable accuracy.