Grid-based k-Nearest Neighbor Approach for Process Monitoring with Large Size Data

This paper presents an algorithmic approach that integrates data mining principles with control chart techniques to detect deviations from standard values within a multivariate dataset. Recently, research has focused on methods for calculating outlier scores based on the k-nearest neighbors (kNN) paradigm. However, the practical utility of kNN-based methods is limited due to the computational complexities inherent in the kNN algorithm, which restrict its applicability to large datasets. The main aim of this research is to propose a new control chart framework that utilizes a grid-based kNN algorithm to reduce the computational effort involved in identifying the k nearest neighbors. To validate the effectiveness of this methodological innovation, extensive experiments were conducted in various experimental settings. The empirical results from these experiments demonstrate significant efficiency gains, as the proposed method considerably reduces the computation time required for analysis while maintaining a level of precision and reliability that is both predictable and acceptable in the context of anomaly detection and control charting.