Twice Ant Colony Clustering Algorithm In Kernel Space And Its Improvement

Cluster analysis is a important content in data mining.It is a important means for us obtain knowledge from mass of data. Ant Colony Algorithm begins with the1990s of last century. Subsequently, bionic intelligence of ant colony have made better result in a lot of fields,Indeed,NP difficult problem.As a new meta-heuristic algorithm, ant colony algorithm has simplicity, extensive adaptability, robusticity; It can calculate distributedly, easy to expand, easy to bind with other algorithms.This text research twice ant colony clustering and it’s matching algorithm, relate to Acluster ant colony clustering algorithm,kernel function, compress algorithm.Since traditional clustering algorithm has unsatisfactory result while dealing with the complicated characteristic data,we have proposed a series of solutions.Because ordinary Euclid distance is difficult to distinguish cluster object,we propose nuclear function mapping method.Original space mapped to high dimension feature space using gauss radial basic function,get eigenvalue and clustering with this eigenvalue. We give a new method to calculate parameter in kernel function.Since result of first Acluster ant colony algorithm have too much clusters,we propose twice clustering.Twice clustering compress big space and object to small space,we propose mosaic algorithm clustering after compress,then we get amount near reality.Since simple mosaic method have phenomena of lose at edge,we bring forward denseness mosaic algorithm,enhance sampling density,reduce lose of information in compress process.Experiment validate that algorithm which we proposed get preferable result.