一种基于近似类抽样的组合聚类方法

he time complexity of k-means clustering method and fuzzy c-means clustering method can be accepted when used in data mining of huge amounts of data sets. But these two methods are sensitive to initialization. When these two methods make an appropriate initialization, they can get a satisfied clustering result. When these two methods make an improper initialization, they can not often get an acceptable clustering result. The k-medoids substitution clustering method based on the idea of simplex method has more good clustering effect and less sensitivity to an initialized medoids set than k-means and k-medoids, when clustering those sets of date points with some similar-size clusters. But its time complexity is too high, so can not be used in huge amounts of data sets. In order to solve their shortcomings, a combination clustering method based on sampling using approximate aggregation is presented naturally. Usually, this method still needs O(n*n*m).