Canonical PSO Based K-Means Clustering Approach for Real Datasets

Area of Science:

• Data Mining and Machine Learning
• Computational Intelligence
• Applied Statistics

Background:

• Clustering is a vital unsupervised data mining technique with broad applications.
• Evaluating clustering results, specifically cluster compactness and separability, is crucial.
• Cluster validity measures are essential for assessing the quality of clustering outcomes.

Purpose of the Study:

• To propose a novel Canonical Particle Swarm Optimization (PSO)-based K-means clustering algorithm.
• To analyze the impact of intercluster and intracluster validity indices.
• To compare the performance of various clustering algorithms on diverse real-world datasets.

Main Methods:

• Implementation of a Canonical PSO-based K-means algorithm.
• Application and analysis of key cluster validity indices (intercluster, intracluster).
• Comparative evaluation using K-means, Canonical PSO-K-means, simple PSO-K-means, DBSCAN, and Hierarchical clustering.

Main Results:

• The study evaluates algorithm performance on air pollution, wholesale customer, wine, and vehicle datasets.
• Mathematical and graphical representations illustrate algorithm behavior with validity indices.
• Identification of the most suitable algorithms for achieving compact clusters on specific datasets.

Conclusions:

• The Canonical PSO-based K-means algorithm shows promise for enhancing clustering tasks.
• The selection of appropriate validity indices significantly influences clustering performance.
• This research provides insights into algorithm selection for effective cluster analysis in real-world applications.