|
Abstract:
|
Hierarchical clustering algorithms are one of the most frequently used approaches in unsupervised learning. But, their applications in many fields have encountered a number of problems, such as the sensitivity of outliers and early stage error. Also, these algorithms cause a large clusters problem when there are very similar clusters with low variance than other clusters. In the article, we propose hierarchical clustering algorithms with unimodal test to settle these problems. When two groups are merged, we try unimodal test for two groups. If data of two groups are unimodal, two groups are merged. Otherwise, two groups are not merged. The proposed method can settle the problems and select the number of clusters K automatically. The performance of the proposed method is examined using both simulated and real data sets.
|
