Combining hierarchical clustering methods

The existence of Internet has caused the creation of big data. Managing this data, which is often unlabeled, is a major challenge. Hierarchical clustering is known as an efficient unsupervised approach for unlabeled data analysis. Hierarchical clustering is a mechanism for grouping data at different scales by creating a dendrogram. In this article, we present a cumulative hierarchical clustering method based on the clustering of clusters and the principal component analysis (PCA) method, which performs the clustering task as a group. The proposed algorithm consists of three main steps. In the first step, a group of individual cumulative hierarchical clustering algorithms are combined to detect relationships between samples and create initial clusters using PCA. In the second step, the initial clusters created by different algorithms are re-clustered to create superclusters. After re-clustering the existing clusters, each sample is assigned to a supercluster with maximum similarity to form the final clusters in the third step. We used the UCI dataset to evaluate the proposed algorithm. Based on the results, the proposed method performs better than other clustering methods such as ENMI and CEGC.

کلیدواژه ها:

Hierarchical clustering ، AHC ، cumulative ، clustering of clusters ، PCA.