|
计算机应用研究 2012
Rough kernel clustering algorithm based on particle swarm optimization
|
Abstract:
According to the disadvantages of the K-means clustering algorithm such as easing to fall into local optimum, can not handle data of boundary objects and non-linear, this paper proposed a rough kernel clustering algorithm based on particle swarm. The sample in the input space was mapped to high-dimensional space by Mercer kernel, so that the sample characteristics which were not shown in the sample space would be appear in the high-dimensional space, and combined with the idea of rough set, by changing the weighting factors of upper and lower approximation dynamically, to cope with the boundary objects efficiently. And reliefF method weighted samples'properties to solve the problem of mixed data clustering. Finally, used an improved particle swarm optimization algorithm to prevent the algorithm into a local optimum. Simulation results show that the algorithm has higher accuracy and shorter convergence time compared with the others improved algorithms, and is verified robustness and stability furtherly, and has some practical value.