기존 k-mode 알고리즘 개선을 통한 범주형 속성 데이터에 대한 효율적인 클러스터링 방법

Abstract

In recent years, as importance of categorical database has increased, various researches about categorical data are performed. Especially, k-mode algorithm which is partitional methods for categorical data as an idea of extending k-means algorithm shows excellent capability in time efficiency. However, similarity calculation through simple comparison between random-choice and mode about initial points mode shows low accuracy about result of clustering. In this thesis, the selecting method of initial points, which is not random-choice about initial points but based on frequency of data attribution has been improved. Also accuracy of clustering results has improved by giving weights which it follow in frequency.