Dynamics Under Clustering Algorithm

Along with the computer technology, the database technology and the network technology rapid development, we have been placed into a data-explosion Time. We need some new and powerful data analysis methods and technologies urgently to solve the embarrassing situation of "Data Rich & Information Poor".The Data Mining means to find the hidden, unknown, novel information with application value from the large-scale databases or the data warehouses. It related with the Database, the Artificial Intelligence, the Machine Learning, Statistics, the High-Performance Computation and so on the multitudinous domain theories and the technical research results. Clustering Analysis already became an extremely active research area in the Data Mining. Most important clustering algorithms have been divided into partition-based method, hierarchy-based method, density-based method, grid-based method, model-based method and so on, or these methods combinations and its improvement at present. These algorithms are almost based on the distance (or density) between the two objects that is a static clustering standard.Introducing the background knowledge of the other subjects provide us some new ideas on clustering standards. Based on the predecessor research results, we placed the data objects into the dynamics knowledge of physics background. This article abstracts the data objects as the physical particles those are attracted by other particles and moving abiding by the Newton's Law of Gravitation and the Newton's Law of Motion. Following the research clue of the interaction gravity of the physical particlesâ†'the changes of the particles displacementâ†'revising the speed and the place through learning constantly, we go on with the further research and discussion on the clustering algorithms.We propose two clustering algorithms HGBCA(a clustering algorithm based on the hierarchy & gravitation) and GGBCA(a clustering algorithm based on the grid & gravitation).The dense objects have a higher priority on clustering, so that it can get better effect than the only clustering standard of distance-related algorithms. Through comparing and inspecting on the displacement changing of particles in physical space, we propose MGBCA (a clustering algorithm based on the motion & gravitation). Regard displacement as an important clustering standard, data objects clustering process acts as the particles moving to the cluster centers .Finally we introduced a clustering algorithm which can study, moving with fuzzy intelligence——FPSO(the fuzzy particle swarms Optimization).The proposal clustering algorithms improve the clustering effect better than those algorithms whose clustering standard just related with distance, thus lead to the isolate, static gather process before. The new algorithms transform the gather process into a related, moving and intelligence process that improved through the study unceasingly. The performance analysis and the results of contrast experiment showed that the new algorithms have enhanced clustering effect and the speed of carrying out. The results reflect the natural relation essence of data objects, improve the clustering quality.