XCSc: a novel approach to clustering with extended classifier system.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Mining

Background:

• Clustering complex and noisy datasets presents significant challenges in data mining.
• Existing methods often struggle with rule redundancy and overlapping cluster boundaries.
• eXtend Classifier Systems (XCS) offer a framework for rule-based learning but require adaptation for effective clustering.

Purpose of the Study:

• To propose and evaluate XCSc, a novel clustering approach based on eXtend Classifier Systems (XCS).
• To enhance the performance of XCS for clustering noisy and complex datasets.
• To improve upon existing clustering algorithms in terms of accuracy and stability.

Main Methods:

• XCSc integrates three core processes: rule population evolution, rule compaction, and rule merging.
• A modified XCS learning mechanism with an accelerated learning method is employed.
• A novel agglomerative hierarchical rule merging algorithm, based on graph modeling, is utilized.
• The approach was compared against the CHAMELEON benchmark algorithm on challenging datasets.

Main Results:

• The proposed XCSc approach demonstrated superior performance compared to the CHAMELEON algorithm.
• XCSc achieved a higher successful rate in clustering complex and noisy data.
• The method exhibited good stability across various challenging datasets.

Conclusions:

• XCSc offers an effective and robust solution for clustering noisy and complex data.
• The novel rule merging process effectively handles overlapping rules between clusters.
• XCSc represents a significant advancement in applying classifier systems to data clustering problems.