一种新的模糊聚类有效性指标

摘要

构造了一个描述数据集模糊划分结果是否清晰的度量.通过综合考虑数据集划分结果的清晰度、紧致性、分离度等因素,得到一个判别模糊聚类的最佳聚类数的有效性指标函数.这个有效性指标函数兼顾到数据集的模糊划分和数据集的几何结构特性,提高了判别结果的准确度.针对人工数据集与真实数据集进行了仿真试验,并与部分已有指标函数的试验结果进行比较,结果表明:所给出的有效性指标函数能够准确地判别试验中所使用的人工数据集与经典真实数据集的聚类数,并且误判率较低.%A measurement is constructed for the clarity of fuzzy partitioning results of data.After that, a validity index function on the best cluster number is obtained for fuzzy clustering by taking clarity, compactness, and separating degree into account.This validity index function gives consideration not only on fuzzy partitioning but also on the properties of geometrical structure of the dataset.Thus, the accuracy of the evaluation is improved.Simulations are carried out for both artificial and real datasets.The simulation results are compared with those results given by some existing validity indices.Simulation results indicate that the validity index function can figure out the best clusters numbers of the tested artificial and real datasets.Furthermore, the error rate is low.