Enhancing Fuzzy C-Means Performance for Low Fuzzifier Values through Ordered Weighted Aggregation

Abstract

A prominent fuzzy clustering technique is the Fuzzy C-Means (FCM) in which the fuzzifier value m has a substantial effect on the algorithm’s performance. Standard FCM may become unstable in this environment even though smaller values of m result in more refined partitions. Ordered Weighted Aggregation Fuzzy C-Means (OWA-FCM), which integrates an Ordered Weighted Aggregation (OWA) operator into the membership update stage, is our approach to address this problem. An orness parameter (p₀) controls the binomial distribution which is used to calculate the OWA operator’s weights. For various values of m and p₀, extensive experimentation on benchmark datasets were performed. Obtained results clearly indicate that, for m ≳ 2.0, OWA-FCM behaves as regular FCM. On the other hand, OWA-FCM generates more accurate and consistent clustering for lower fuzzifier values (m ≲ 1.8), particularly m ≤ 1.4, leading to better ARI scores and crisper partitions. In addition, throughout a broad range of orness values (0.001 ≤ p₀ ≤ 0.999), the final clustering results remain unchanged, suggesting that the OWA-based aggregating mechanism, rather than the particular selection of p₀, is accountable for the improvement. These results indicate that when more precise cluster assignments are required, OWA-FCM is a reliable substitute for conventional, standard FCM.

Keywords

Article Information