优化K均值随机初始中点的改进算法

Three Improved Algorithms for Optimizing of Randomly-initiated K-means Midpoints

针对传统K均值随机产生的初始聚类中心的方式提出最近邻K均值、极远邻K均值和自适应K均值3种优化算法。最近邻K均值是通过寻找多维空间下欧氏σ邻近点的方式确定K群;而极远邻K均值是极远σ邻判定确定法;自适应K均值是将数据集确定到矩阵中,对矩阵做归一化、二元化处理后,计算各向量间的相异度来修正确定初始中心点的加权欧氏距离。3种优化算法改善了原始K均值算法,提高了算法的稳定性和精确度,而且它们各自适用于不同的应用空间。

Aiming at traditional K-means which randomly generating initial clustering centers,the N2-K-means,F2-K-means and SA-K-means center point optimization algorithms were proposed,in which,the N2-K-means determines group K by looking for Euclideannearby points in multidimensional space;and F2-K-means employs a very far neighbor method;and the SA-K-means converts the data sets from the separate data into matrix,and then has them normalized and dualized to calculate the dissimilarity between every vector so as to modify Euclidean distance of each initial center.This three algorithms suitable for different spaces can improve the traditional K-means algorithm and promote its stability and accuracy.