The Core Cluster-Based Subspace Weighted Clustering Ensemble

Abstract

In recent years, the Internet of Things (IoT) technology has developed rapidly and is widely used in various fields. It is of great research significance to uncover underlying patterns and insights from the high-dimensional data of IoT, to excavate valuable information to guide people’s production and life. Clustering can explore the natural cluster structure of the data, which is conducive to further understanding of the data, and is an essential preprocessing step for data analysis. However, clustering is highly dependent on the data. In order to reduce the complexity of the model, reduce the computational cost, and obtain a more robust clustering solution, we combine subspace clustering and ensemble learning to propose a novel subspace weighted clustering ensemble framework for high-dimensional data. The proposed framework first combines random feature selection and unsupervised feature selection to generate a set of base subspaces. Clustering is performed on each base subspace to achieve a set of subspace clustering solutions that generate a set of adaptive core clusters. The size of the core cluster is between the sample and the cluster. In the ensemble process, the core clusters are viewed as the basic unit, and the stability of the cluster is evaluated by measuring the distance between the core cluster pairs, and the similarity between the core clusters and the clusters in the base subspace, and then weighting the subspace clustering solution. Under this framework, we propose four subspace ensemble approaches based on core cluster to improve the accuracy of consensus clustering solutions. Extensive experiments are conducted on multiple real-world high-dimensional datasets, demonstrating that the proposed framework can process high-dimensional data for the IoT, and the proposed subspace clustering ensemble approaches are superior to the state-of-the-art clustering approaches.