Based On Entropy And Distance Weighted Multi-angle Fuzzy Clustering

With the progress of the times and the development of science and technology,the amount,types and forms of data that need to be processed in life are increasing rapidly,and how to extract useful knowledge from massive data has become more difficult.In such a severe situation,it is urgent to improve the ability of data mining.The clustering method has always been a very important data mining method,but the traditional clustering algorithm is mostly single-angle clustering algorithm,processing single-angle data,single-angle data is a large number of the same kind of data in the same form,which is completely inconsistent with the actual demand.Although the single-perspective clustering algorithm can cluster the data of each perspective separately,and finally integrate the clustering results through the integration method,this method does not take into account the influence of different perspectives,and the results will be quite different from the real results.Therefore,single-angle clustering is not suitable for multi-angle data.The same data will get different measurement results from different angles or fields,and the measurement results from different angles or fields can be analyzed by clustering algorithm alone.Such data is called multiperspective data.In the face of multi-angle data that cannot be processed,the multi-angle clustering algorithm is created on the basis of the single-angle clustering algorithm.The multi-angle clustering algorithm deals with the measurement results of each angle respectively.Under the principle of complementarity and consistency,the relationship and interaction between perspectives are established to successfully solve the defects of single-angle clustering.However,there are still some problems to be solved in the multi-perspective clustering algorithm.Based on FCM and PCM clustering,this paper designs corresponding solutions to the problems existing in the multi-perspective FCM clustering algorithm and multi-perspective PCM clustering algorithm.Firstly,fuzzy c-means clustering algorithm for multiple points of view of the fuzzy index m physical meaning is not clear problem,put forward different perspective space of the harvard entropy based on the principles,the nature of the entropy describes the uncertainty of a random variable and fuzzy membership degree to describe the uncertainty of the sample points belong to the nature of the close,the entropy theory is introduced into the fuzzy clustering algorithm,the objective function to obtain the optimal solution,the perspective of harvard minimum entropy.Secondly,aiming at the problem that the default attributes in the multi-perspective fuzzy c-means clustering algorithm are the same,a distance weighting strategy based on rough set is proposed.In the preprocessing,the rough set algorithm is used to set a weight for each attribute,and the weight of the attribute should be considered when calculating the distance between the attributes.Finally,fuzzy c-means clustering algorithm for multiple points of view in the default view of the same problem,put forward with adaptive weighted characteristic of multiple points of view,introducing weight coefficient,and the information entropy,said according to the principle of maximum entropy,make the objective function in optimal solutions,highlights the most clustering effect,from the point of view of reducing clustering characteristics in algorithm on the basis of poor,by raising the perspectives of the fuzzy c-means clustering effect of clustering.To solve the problem of uniform default perspective weights in multiperspective PCM clustering algorithm,an inter-perspective weighting method is proposed,which assigns reasonable weight to all perspectives and the sum of all perspective weights is one.In view of the problem that all attribute weights in the multi-perspective PCM clustering algorithm are the same by default,an intraperspective attribute weighting method is proposed,which gives the attribute with good clustering quality a large weight and the attribute with poor clustering quality a small weight.Experiments on data sets in UCI database show that this algorithm has better clustering performance for processing multi-angle data.