Boosting k-means clustering with symbiotic organisms search for automatic clustering problems

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• K-Means clustering is a widely used unsupervised learning algorithm for data partitioning.
• K-Means is sensitive to initial centroid placement, potentially leading to local optima and suboptimal clustering.
• Existing hybrid methods often require extensive parameter tuning for metaheuristics.

Purpose of the Study:

• To propose a novel hybrid clustering algorithm, SOS-KMeans, combining Symbiotic Organisms Search (SOS) with K-Means.
• To leverage SOS as a parameter-free metaheuristic for generating optimal initial centroids for K-Means.
• To improve the performance and robustness of automatic clustering.

Main Methods:

• Developed a hybrid clustering approach integrating the Symbiotic Organisms Search (SOS) algorithm with the K-Means algorithm.
• Utilized SOS as a global search metaheuristic to determine optimal initial cluster centroids for K-Means.
• Evaluated the proposed SOS-KMeans algorithm against classical K-Means, classical SOS, and other hybrid clustering methods.

Main Results:

• The SOS-KMeans algorithm demonstrated improved performance in automatic clustering tasks.
• Comparative analysis on eleven UCI Machine Learning Repository datasets and one artificial dataset showed superior results.
• The proposed hybrid method outperformed standard K-Means and other existing hybrid clustering approaches.

Conclusions:

• The hybrid SOS-KMeans algorithm effectively addresses the limitations of the standard K-Means algorithm.
• SOS-KMeans offers a robust and efficient solution for automatic clustering with minimal parameter tuning.
• This approach represents a significant advancement in metaheuristic-based clustering techniques.