Ensemble Clustering with Novel Weighting Strategy

The target of ensemble clustering is to improve the accuracy of clustering by integrating multiple clustering results and solve the problem of scalability existed in traditional and single clustering algorithms. In recent years, ensemble clustering has attracted increasing attention due to its remarkable achievements. However, the limitation of most existing ensemble clustering approaches is that all base clusterings are treated equally without considering the validity of them. Some ensemble clustering algorithms are aware of using weighting strategy but also ignoring the negative impact of base clusterings with poor performance. In this paper, we propose an ensemble clustering method based on a novel weighting strategy. Specifically, the validity of each base clustering is measured by the optimal matching score between the base clustering and the whole to obtain the corresponding weight. Then, the weights of base clusterings which have negative contribution are further adjusted to get the final weight vector. Subsequently, a weighted co-association matrix is constructed to serve as the ensemble matrix and a hierarchical clustering algorithm is applied to it to generate the final result. Experimental results on different types of real-world datasets show the superiority of proposed methods.