Performance Analysis of a Novel IT2 FCM Algorithm

Abstract

In this paper, we propose a novel interval type-2 (IT2) fuzzy clustering algorithm by incorporating a speed up type reduction algorithm. In order to illustrate our proposed method, embedded lines and planes that are associated with the IT2 fuzzy membership functions (MFs) are confined to 2-dimensional (2-D) space for visualization purposes. The original IT2 fuzzy C-means (FCM) algorithm uses the Karnik-Mendel (KM) algorithm as a part of its type reduction procedure where computation of the centroid is achieved by iterating each dimension of the pattern sets separately. This ignores the possible correlation among the multiple dimensions and can result in high computational complexity. Our proposed algorithm considers multidimensional pattern sets jointly and estimates the centroid at comparable costs. Finally, experiments are performed on several pattern sets to show the validity of our proposed method.