Rough Hypercuboid Based Generalized and Robust IT2 Fuzzy C-Means Algorithm.

Area of Science:

• Pattern Recognition
• Machine Learning
• Data Mining

Background:

• Clustering algorithms aim to identify natural groupings within datasets.
• Real-life data often presents uncertainty, posing challenges for traditional clustering methods.
• Existing fuzzy clustering algorithms struggle with parameter selection and uncertainty handling.

Purpose of the Study:

• To propose a novel clustering algorithm, rough hypercuboid-based interval type-2 fuzzy c-means (RIT2FCM), designed to handle data uncertainty.
• To integrate the strengths of rough hypercuboid approach, c-means, and interval type-2 fuzzy sets for robust clustering.
• To address the difficulty in determining appropriate fuzzifier values in rough-fuzzy clustering.

Main Methods:

• The algorithm utilizes the hypercuboid equivalence partition matrix (HEM) to implicitly define cluster boundaries without thresholds.
• Interval type-2 fuzzy sets are employed to manage parameter uncertainty.
• Analytical formulation for convergence analysis and a theoretical bound for the fuzzifier are developed.

Main Results:

• The RIT2FCM algorithm demonstrates superior performance compared to existing state-of-the-art c-means algorithms in 92.59% of cases.
• Performance was evaluated using cluster validity and classification rate indices on diverse real-life datasets.
• The proposed method achieves better results with reduced computation time.

Conclusions:

• RIT2FCM effectively addresses uncertainty in datasets, outperforming traditional clustering methods.
• The algorithm offers improved accuracy and efficiency in pattern recognition tasks.
• Implicit definition of cluster regions and interval-valued fuzzifiers contribute to its robustness.