Evaluating the Applicability of Self-Tuning Algorithms in Diverse Clustering Scenarios: An ST-PFCM Case Study
Szilárd Sipos, Mirtill-Boglárka Naghi, Levente Kovács · 2025
Fuzzy$c$-means clustering has a well-established history as a foundational method in unsupervised learning. Over time, several advancements have been introduced, including the development of possibilistic$c$-means (PCM), which assigns possibilistic values representing the degree of possibility that a data point belongs to a specific cluster. Building upon these concepts, the possibilistic fuzzy$c$-means (PFCM) algorithm was proposed, integrating the strengths of both FCM and PCM through a linear combination of their principles. While PFCM demonstrated significant improvements in clustering performance, it introduced the drawback of requiring manual tuning of additional parameters compared to its predecessors, thereby increasing computational complexity and effort in obtaining optimal results. Recently, a novel approach termed self-tuning possibilistic fuzzy$c$-means (ST-PFCM) has been introduced in the literature. This self-tuning algorithm addresses the parametertuning challenge by automatically adjusting its parameters, representing a promising new direction in the field. However, the initial studies validating ST-PFCM primarily focused on toy and benchmark datasets, limiting its demonstrated applicability to real-world scenarios. The present study aims to extend the evaluation of ST-PFCM by applying it to real-life datasets. This investigation seeks to further assess the algorithm's potential and practical utility in diverse, real-world contexts.