Core-mergence clustering method and joint-feature judging model for anomaly detection
Jinyu Yang, Shaowu Lu, Wanqing Li · Journal of Control and Decision · 2024
In the manufacturing process, there is an explosive growth in the amount of data, which is directly unsuitable for anomaly detection. To solve the problem, this paper proposes a core-mergence clustering method and a joint-feature judging model. First, to obtain the relative positional relationships of the clustering centers, the initial data cores and the boundary points for each cluster are calculated by a local density estimation method, thereby obtaining the initial clustering radius. Then, based on a density-based weighting method, the data cores are adaptively merged for further refining the clustering centers, as well as the clustering radius are synchronously updated to form the corresponding regular hypersphere. Finally, by analyzing the overlapping degree between different clusters, the joint-feature model using a new similarity metric is employed to reclassify the suspicious points, forming the judging boundaries for the overlapping clusters. The proposed method is verified by both simulated and experimental datasets.