Research On Cluster Ensemble Methods Node-Important-Based

As an important research direction in machine learning,ensemble learning is widely used in military,scientific research,social life and other aspects.Ensemble learning is a machine learning method that uses a series of different learners to integrate the results to solve the same problem.It has been widely used in many fields,such as Internet communication,satellite tracking,earthquake monitoring,human gene bank,speech recognition,medical intelligent diagnosis and so on.In the early stage of the ensemble learning research,it mainly focuses on the supervised learning.As a kind of unsupervised learning,clustering ensemble has attracted more and more attention just in recent years and has become a hot topic in the field of machine learning.By using the ensemble learning technique,a new and more accurate result can be obtained by clustering and integrating multiple base clusterings of a data set.At present,the clustering ensemble mainly focus two issues: the first is the generation of the base clusterings;the other is the design of consensus function.The existing clustering algorithm mostly regard the base clusterings as a categorical dataset,and does the secondary clustering with the base clusterings,but without the combination of attribute values of the data point itself,and ignoring the original dataset,thus the final clustering information is incomplete.Based on this condition,we carry on the following researches according to the design of the consensus function in the ensemble clustering.The main contents as follows:(1)Based on the existing consensus function,a Cate-NIR algorithm for categorical data is proposed.This algorithm uses the Node Importance Representative to design the consensus function,and designs the experiments on the UCI categorical data to prove the universality of the algorithm.(2)Based on the idea of Cate-NIR algorithm,aiming at the particularity of the numeric data to modify,and we propose a Num-NIR algorithm for numerical data,which verifies the applicability of the algorithm on the general dataset.The above work obviously expands the researching fields of clustering ensemble algorithm and provides a new research direction for the design of consensus function.Meanwhile,it also lays the foundation of the researching of cluster ensemble algorithm.I believe that the continuous progress of such algorithms can solve more and more practical problems.