A black-winged kite improved fuzzy clustering handling imbalanced uncertain data

Area of Science:

• Data Mining
• Machine Learning
• Artificial Intelligence

Background:

• Clustering uncertain data is challenging due to imbalance, where minority clusters are often overshadowed.
• Existing techniques struggle with initialization bias and inadequate similarity modeling for uncertain and imbalanced datasets.

Purpose of the Study:

• To introduce the Black-winged Kite Improved Fuzzy clustering for probability density Functions (BKIFF) algorithm.
• To address the limitations of existing methods in handling uncertain and imbalanced data clustering.

Main Methods:

• BKIFF combines an optimization-based initialization strategy with an enhanced fuzzy clustering framework.
• It incorporates Hellinger distance for reliable similarity modeling of probability density functions (pdfs).
• Improved membership updating, prototype estimation, and Black-winged Kite Optimization (BKO) enhance clustering stability and convergence.

Main Results:

• BKIFF consistently outperforms baseline methods (FCF, KMEANS, Self-Updating) on synthetic and real-world datasets.
• Achieves significant improvements in Adjusted Rand Index (ARI) and Normalized Mutual Information (NMI), especially in imbalanced scenarios.
• Reduces computational time by 95-99% compared to baseline methods.

Conclusions:

• BKIFF demonstrates superior performance in clustering uncertain and imbalanced data.
• Offers new possibilities for applications in medical diagnostics, ecological analysis, and high-dimensional uncertain data mining.