Research Swarm Intelligence Optimization Clustering Algorithm

With the fast development of information technology, the amount of data in the realistic society also presents an explosive growth. People are eager to seek internal relations and find the hidden knowledge from the large and complex database. Clustering analysis is used to identify the data subjects and find the internal relations in the dataset. It has a very important role in the field of data mining. According to analyzing and summarizing the traditional clustering algorithm’s problems and defects, the swarm intelligence algorithms is applied to optimize the clustering algorithm by combining with the characteristics and advantages of swarm intelligence algorithm.The main content of this paper are as follows:Firstly, the traditional K-Means algorithm is over-dependent on the choice of the initial cluster centers during the clustering process. Meanwhile, due to the lack of global search capability, it is difficult to get the accurate cluster centers. Fish-school algorithm shows good parallelism and global search feature in solving optimization problems, but may falls into local optimal solution because of the artificial parameters. According to their characteristics, apply the fish-school algorithm with adaptive step to the clustering problems, then immunity-vaccination mechanism is combined to strengthen the search performance of the algorithm for the exact solution. The experimental analysis and comparison results on UCI datasets show that the algorithm has better validity and stability.Secondly, traditional partition clustering method has the problem of over-reliance on the initial cluster centers and the method is prone to fall into local optimum. So an improved partition clustering method based on the firefly algorithm is proposed. The method considers an firefly as a set of cluster centers and class cohesion is regarded as brightness of the firefly. Then find the optimal clustering center by the fireflies attracting each other. In the process of optimization, randomly distributed firefly populations is used to overcome the problem of over-reliance on the initial cluster centers and adaptive step strategy is adopted to strengthen the ability to find the exact solution of the algorithm. In order to prevent the algorithm from local optimum for population concentration, the niching technology was introduced to improve the diversity of the fireflies’ population. Experiments on Iris datasets and Cancer datasets indicate that the algorithm is improved at clustering precision and stability.Finally, it has the important biological significance to use the clustering method to predict and identify protein function module through analyzing the protein interaction relationship in the Protein-protein interaction (PPI) network. In recent years, some new algorithms are proposed but the clustering result still need to be improved. A PPI network clustering model based on the mechanism of the artificial immune system was proposed. In which the set of cluster centers were regarded as antigens and the neighbor nodes as antibodies. The antibodies were regarded as the memory cells divided into clusters by calculating the affinity between the antibodies and antigens. Then select excellent antibodies as vaccines and try to inject the vaccines into clustering modules and update it. Afterward keep the memory cells updated after comparing the fitness of the modules before injecting. The simulation experiment on PPI datasets showed that the f-measure value of the new algorithm gets improved.