An Effective Clustering Algorithm Using Adaptive Neighborhood and Border Peeling Method.

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Traditional clustering algorithms necessitate pre-defined neighborhood parameters and the number of clusters.
• Optimal parameter selection in conventional methods is data-distribution dependent, requiring domain expertise.
• These limitations hinder the generalizability and ease of use of existing clustering techniques.

Purpose of the Study:

• To develop an effective clustering algorithm that bypasses the need for neighborhood parameter and cluster number selection.
• To provide a robust clustering solution applicable to datasets with varying shapes and distributions.
• To enhance the accessibility and applicability of clustering methods in data analysis.

Main Methods:

• An adaptive neighborhood approach is employed, iterating to a stable state to derive neighborhood information.
• The algorithm identifies and processes boundary points based on the adaptively determined neighborhood characteristics.
• Core points are utilized to form and define the final data clusters.

Main Results:

• The proposed algorithm successfully clusters data without requiring manual setting of neighborhood parameters or the number of clusters.
• Extensive comparative experiments on diverse datasets demonstrated satisfactory and robust clustering outcomes.
• The method shows effectiveness across datasets of varying sizes and distributions.

Conclusions:

• The adaptive neighborhood clustering algorithm offers a significant improvement over traditional methods by automating parameter selection.
• This approach enhances the practical utility of clustering for a wider range of data analysis tasks.
• The algorithm provides reliable clustering results irrespective of data characteristics or prior user knowledge.