Study On An Improved Algorithm And Simulation Experiment Based On Ant Colony Clustering

Swarm Intelligence is more and more widely noticed for its interesting properties such as flexibility, robustness, decentralization and self-organization.Clustering based on Ant Colony Algorithm is a kind of data mining algorithm. It emerges out of researches on ant collective behavior. Lumer and Faieta extended the Basic Model which was put forward by Deneubourg to apply it into clustering analysis area. But LF Algorithm has some disadvantage, such as it isn't able to disjoin different clusters automatically which are accidentally superposed in a local area, making the result of clustering to have the low purity, when badly hurting the precision.In order to overcome the disadvantage of LF algorithm, by means of combining Fuzzy Clustering Algorithm, an improved algorithm based on ant colony clustering is put forward. While tracing back to Deneubourg's basic model, the new algorithm introduces new concepts such as similarity factor and dissimilarity factor to change the calculation of the perceived fraction f, as to influence the value of Pick-up Probability and Dropping Probability. The value of similarity factor is decided by the size of the equivalence class which the selected object belongs to in the local area while the value of similarity factor is decided the size of the largest equivalence class among which the selected object doesn't belong to at site. According to the method, ants have some discernment which can in the mind compartmentalize objects around them to rough clusters at site, then to decide next step.For this method, even if two different clustering kernels are produced in a local area, with the imbalance of the clustering paces, the preponderant clustering kernel will exclude the other clustering kernel.A data mining system based on Ant Colony Clustering Algorithm is designed. The system has data preprocessing, data transformation and data mining functions. The system visualizes the process of clustering. After preprocessing, transformating data objects, the specified data with three dimension attributes are gotten. On the interface of the software, the three dimension attributes of objects are identified by the value of color. In experiment of the simulation system, testing result given by improved algorithm appears to have better quality than LF algorithm by F-measure evaluation standard.