Interval Type 2 퍼지접근방법을 이용한 Possibilistic C-means

Abstract

Possibilistic approach to C-means(PCM)은 Fuzzy C-Means(FCM) 의 단점을 극복하기 위해 제안되었지만 초기 클러스터 prototype의 값에 민감하고, 상대적으로 가까이에 위치한 prototype들을 형성하는 패턴들의 경우에는 최종 prototype의 위치가 겹치는(동일한) 결과가 나올 수도 있다는 단점이 있다. 이와 같은 단점들을 극복하기 위해서 본 논문에서는 PCM 알고리즘에 Interval Type-2 퍼지접근 방법을 적용하였다. 그 결과 본 논문에서 제안하는 알고리즘은 PCM 보다 나은 성능을 보이며, 특히 상대적으로 가까이에 위치한 클러스터들을 클러스터링 할 때 우수한 성능을 보이고 있다.; The conventional possibilistic approach to C-means (PCM) has been proposed to overcome some of the drawbacks associated with the Fuzzy C-means (FCM) for partitioning pattern data. However, the performance of PCM is known to be sensitive to the initial cluster prototypes and to patterns that have relatively short distances between the prototypes. Hence, this can lead to some final prototypes to be approximately identical that are unrepresentative of the pattern data. To overcome these drawbacks, we propose an interval type-2 fuzzy approach to PCM by considering uncertainty in the fuzzifier parameter m and in the “bandwidth” η_(i) in PCM-based methods. As a result, the proposed algorithms outperform the conventional PCM algorithm especially for pattern data that have relatively short distances between the prototypes. This is greatly due to the extra degree of freedom where two values of fuzzifier m or bandwidth η_(i) is used. Several experimental results are given to show the validity of the proposed algorithms.