Research On The Clustering Ensemble Algorithm Based On The Mahalanobis Distance Metric

Clustering analysis has received a lot of attention in recent years as an important part of machine learning,pattern recognition,and data mining.Clustering analysis by grouping them in unlabeled data and mining the natural division and structure of the data.However,most existing clustering analysis algorithms are heavily reliant on datasets,and a single clustering algorithm cannot handle datasets with varying data structures effectively.The researcher proposed the concept of clustering ensemble algorithm to solve the above problems by drawing on the idea of ensemble learning.Multiple clustering results are fused to obtain a more robust and stable clustering result.The quality of base clustering in clustering ensemble algorithms plays a crucial role in the clustering ensemble results.Therefore,in this paper,we conducted a study related to the quality of base clustering,which are mainly studied as follows:(1)Metric learning guided weighted clustering ensemble is proposed,which incorporates the Mahalanobis distance metric learning and the base clustering weight learning and aims to improve the performance of the clustering ensemble algorithm.The clustering ensemble result of this algorithm in each round is based on the base clustering weights calculated in the previous round,and constructs pairwise constraints based on the ensemble results,thus learning the Mahalanobis distance metric,so that samples within the same clustering are closer and samples in different clustering are more separated.Finally,in the new projection space,the algorithm learns new weights for each base clustering.(2)Clustering ensemble algorithm with regression cluster center is proposed,which mining the prior information through the previous round of clustering ensemble results and performing supervised metric learning based on this prior information to better characterize the similarity between samples and assist in base clustering generation,thus improving the base clustering performance.By guiding the base clustering generation process and the clustering ensemble process iteratively with each other,the base clustering continuously converges and gradually improves the clustering ensemble performance.