基于自适应权重的模糊C-均值聚类算法

Fuzzy C-means clustering based on self-adaptive weight

针对模糊C-均值聚类算法过度依赖初始聚类中心的选取,从而易受孤立点和样本分布不均衡的影响而陷入局部最优状态的不足,提出一种基于自适应权重的模糊C-均值聚类算法。该算法采用高斯距离比例表示权重,在每一次迭代过程中,根据当前数据的聚类划分情况,动态计算每个样本对于类的权重,降低了算法对初始聚类中心的依赖,减弱了孤立点和样本分布不均衡的影响。实验结果表明,该算法是一种较优的聚类算法,具有更好的健壮性和聚类效果。

Due to fuzzy C-means clustering algorithm rely heavily on randomly select C clustering centers, so outlier and uneven distribution of the samples easily influenced and made it easy to fall into the local optimum states. Therefore, this paper proposed an improved fuzzy C-means clustering algorithm based on self-adaptive weights. The new method expressed weight by using the Gaussian distance ratio,it computed the weights for every data according to the current clustering state and no more did rely on the initial clustering center, weakened the influence of outlier and uneven distribution of the samples. The experiments indicate that the fuzzy C-means clustering algorithm based on self-adaptive weights is an effective fuzzy clustering algorithm, has more robust and higher clustering accuracy.