Verification of Clustering Accuracy by Applying Direction-Based Method and Data Conversion

Abstract

In this study, the process of clustering into a group with similar characteristics is shown through pattern analysis, and the similarity is examined whether this group has homogeneity. A direction-based method for clustering the time series data is introduced, and the grouping process is carried out through the direction setting by up or down and the logical operations thereafter. The similarity is verified by comparing the parts within the group after clustering. For more effective verification, data conversion, a data homogenization process, is performed. MAD, MSE, MPSE and TS are reviewed as similarity indicators. For data such as time series data, MPSE and TS, which are scale-independent measurements, are recommended.