The Research On Clustering Ensemble Based On Particle Swarm Optimization Algorithm

Clustering ensembles is an open proposition,which aims to solve the limitation of a single clustering algorithm on the diversity of data structures in data partitioning.It obtains consensus clustering by integrating multiple clustering algorithms.Compared with the single clustering algorithm,it can process data more accurately and stably,and has a better tolerance to the structure and diversity of the data.The particle swarm optimization algorithm is a good meta-heuristic algorithm,which is widely used in many fields.Because it only needs to adjust fewer parameters,it runs very efficiently.The core idea is that particles can quickly find the best particle to complete the optimization problem through information interaction with other particles in the population.A clustering ensemble model: Particle Swarm Optimization For Clustering Ensembles(PSOFCE)is proposesed in this thesis,which combines a variety of classic clustering algorithms to extract multi-dimensional features from the original data set to obtain rich clustering results.The initial clustering result set is formed,and the initial clustering result set after the unified processing of Hugarin algorithm is set as the target consensus solution set.The rapid optimization ability of particle swarm is combined with Jaccard similarity coefficient to search for the most consistent target clustering with the current consensus solution set.In the continuous optimization movement of particle swarm,Jaccard similarity coefficient is used as the fitness function of particles to guide particle swarm in the optimization process,when the fitness value no longer changes or the maximum number of iterations is reached,the optimal target cluster is obtained.Based on the proposed unsupervised clustering ensemble model PSOFCE,by selecting different proportions of data in the data set and applying unsupervised clustering ensemble to cluster them,the labeled data is obtained,and these data are used to construct Must-Link and Cannot-Link constraints,and then these constraint information are added to the standard clustering result set,as the supervision information to guide the clustering integration process,Then,Particle Swarm Optimization For semi-supervised Clustering Ensembles model(semi-PSOFCE)is obtained.Finally,this PSOFCE model is evaluated the clustering effect by using a variety of classic clustering effect evaluation methods in this thesis,and it is conducted multi-dimensional comparison experiments with other excellent clustering ensemble models and single clustering algorithms.Based on experiments on public data sets,it is proved that the clustering ensemble model based on particle swarm optimization proposed in this thesis can obtain a better data consensus clustering effect.Then,for semi-supervised clustering ensemble semi-PSOFCE,through the construction of different proportions of constraint information,a clustering ensemble model of different proportions of supervised information is formed.In the comparison of the gradient of supervised information of different proportions and the comparison of its clustering effect with the unsupervised clustering ensemble model,it is proved that the supervised information with paired constraints can effectively improve the clustering performance of the clustering ensemble model.