一种引入参数无需确定聚类数的聚类算法

A clustering algorithm with parameters that no need to determine the clustering number

针对传统k-均值聚类算法的两大缺点,即算法中需要知道确定的聚类数和初始种群选取的随机性,提出了一种新的聚类算法,即基于k-均值聚类算法的无需确定聚类数的聚类算法。这种算法是基于递增思想的聚类算法,最大的特色是无需事先知道聚类数,初始聚类数取1,初始聚类中心为所有数据点的聚类中心,算法中首先设定一个惩罚参数,对于确定的惩罚参数,运算时聚类数逐渐增加,直到收敛,即聚类数不再发生变化,就得到了所需的聚类数以及最终的聚类结果。运用于茶叶分类和各省市平均工资水平分析的2个实验也验证了这种算法的可行性,通过实验可知,这种聚类算法具有较好的全局收敛能力和较高的正确率,稳定性强,收敛速度快。

After analyzing the two shortcomings of the traditional k-means clustering algorithm, namely the need of knowing the number of clusters and the randomness of selecting the initial population, a new clustering algorithm based on k-means clus- tering algorithm with no need to determine the clustering number is proposed. This algorithm is a clustering algorithm based on incremental theory, without having to know the number of clusters. Let the initial number of clusters to be one, and the initial cluster center is the cluster center of all data points. Firstly a penalty parameter should be settled. For a determined penalty pa- rameter, the number of clusters gradually increases until convergence, which means that the number of clusters does not change again, and then the number of cluster and the final result of clustering can be got. The two classification experiments used in the classification of tea and analysis of average salary of the provinces and cities in China verify the feasibility of the algorithm. From the two experiments, it is shown that this algorithm has good global convergence ability, high accuracy, good stability and fast convergence.