Density-Adaptive Fuzzy Clustering with Isolation Kernel

Abstract

Fuzzy clustering algorithms have become essential tools in data analysis, yet they face significant challenges when dealing with complex and real-world datasets. Fuzzy C-Means (FCM) and its variants, such as Kernel FCM, Robust FCM, and Robust Kernel FCM struggle with clusters of varying densities and shapes, often due to their reliance on fixed distance metrics and sensitivity to initial conditions. These limitations underscore the need for a more versatile and adaptive clustering approach. This paper introduces the Isolation Kernel Fuzzy C-Means (IK-FCM) algorithm, a novel method designed to address these shortcomings. IK-FCM innovatively integrates the Isolation Kernel’s density-adaptive similarity measure with FCM’s soft clustering capabilities, enabling more accurate and flexible clustering across diverse data distributions. Through comprehensive evaluations against standard FCM, Kernel FCM, Robust FCM, and Robust Kernel FCM, IK-FCM demonstrates superior clustering quality and accuracy on both synthetic and real datasets, as measured by Normalised Mutual Information (NMI). The algorithm’s effectiveness is further validated through its application to brain tissue segmentation, highlighting its potential in critical real-world tasks. By effectively addressing the limitations of existing fuzzy clustering techniques, IK-FCM represents a significant advancement in unsupervised learning, offering improved accuracy and adaptability across a wide range of data challenges.