Some Affinity Propagation Clustering Algorithms Based On Swarm Intelligence Optimization And Their Applications

AP has been paid much attention by many scholars and widely recognized in various fields.However,the AP still has three problems:1.The important parameters of the AP,the preference and the damping coefficient,need to be manually adjusted,which will increase the time cost of the algorithm and lead to a decrease in the accuracy of the algorithm;2.The traditional AP can only handle the convex data or clustered structured hyper-sphere data,when faced with complex large-scale data-sets,cannot obtain reasonable clustering results;3.AP uses Euclidean as similarity,while Euclidean distance cannot accurately reflect data.The similarities between the three;this article aims at the above three issues,and builds reasonable similarity measures and preference as the goal,conducts in-depth research and proposes different improved algorithms.1.The preference for affinity propagation clustering algorithms cannot be reasonably determined.A semi-supervised affinity propagation based on artificial bee colony is proposed.The algorithm combines artificial bee colony algorithm with affinity propagation algorithm,adjusts similarity matrix with semi-supervised idea,and introduces bee into the iterative process of the algorithm.Because of its strong searching ability,bidirectional biasing parameter space can be searched to obtain the optimal value.Simulation experiments show that the ABC-SAP clustering algorithm is consistent with the expected results and the clustering accuracy is improved.2.In view of the unsatisfactory effect of the affinity propagation clustering algorithm on the complex big data set,this paper proposes a new affinity propagation algorithm based on large scale data set.When the distance of clustering center is determined,the structural similarity is calculated,and then the peak density clustering algorithm is introduced to re-cluster all the clustering centers to make it achieve the best clustering results.3.In order to guide AP clustering more effectively by preference and damping coefficients,a new algorithm is proposed—adaptive step size cuckoo search-based affinity propagation clustering algorithm.Based on the characteristic of the Cuckoo algorithm,this paper applies the Cuckoo algorithm to the AP algorithm to find the best deviation parameter value and damping coefficient value,and according to the validity index Silhouette value,the best clustering is determined.The simulation results tell us that the ASCS-AP clustering algorithm improves the quality and stability of the AP algorithm.4.In order to expand the field of application,this paper applies the ABC-SAP algorithm and the ASCS-AP algorithm to the clustering of the average monthly expenditure data of rural households.It is proved that the clustering result of the new algorithm is basically consistent with the real classification,which shows that the new algorithm can play a role in practical application and provides a basis for the analysis of national economy.