基于数据密集性的自适应K均值初始化方法

摘要

K-means clustering algorithm is widely used in data mining and machine learning region.However the choosing of the initial clustering centroids greatly influences the entire clustering effect.Therefore how to reasonably initialize K-means clustering algorithm becomes an important research orientation.The article proposes an adaptive initialization clustering center choosing method based on data intrinsic den-sity,which is implemented in two stages.Stage one,it provides the definition of data density and chooses,on the basis of data density,the can-didate initialization clustering centroids that meet requirements;stage two,after-process is executed on the chosen candidate initialization cen-troids so that their number conforms to the data class.Experiment proves that the proposed method has the following advantages:1 )it can au-tomatically discover the density of data distribution from datasets and can reasonably find out the initialization clustering centroids;2)it is ro-bust to outliers and noises;3)it reduces the iteration steps of k-means clustering algorithm;4)it is convenient to implement.%K均值聚类算法在数据挖掘、机器学习领域被广泛应用。但其初始聚类中心的选取对整个聚类效果会产生很大的影响，因此，如何合理地初始化K均值聚类算法成为重要的研究方向。提出一种基于数据内在密集性的自适应初始聚类中心选取方法。该方法分为两个过程，第一个过程给出数据密集性的定义，并基于数据密集性选出满足条件的候选初始聚类中心，第二个过程是对选出的候选初始中心进行后处理，使其个数与数据类一致。实验证明，提出的方法有如下优势：1）能够自主发现数据集中数据分布的密集性，并能够合理找出初始聚类中心；2）对离群点和噪声鲁棒；3）减少了K均值聚类算法的迭代步骤；4）易于实现。