A novel density peaks clustering algorithm for automatic selection of clustering centers based on K-nearest neighbors

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Traditional density peak clustering (DPC) algorithms require manual center selection, limiting performance on complex datasets with varying densities or non-convex shapes.
• Existing DPC variants struggle with automatic cluster center identification, necessitating improvements for broader applicability.

Purpose of the Study:

• To propose a novel density peak clustering algorithm (AKDPC) for automatic cluster center selection.
• To enhance clustering accuracy and robustness on diverse and complex datasets.

Main Methods:

• AKDPC classifies samples into core and non-core points based on mutual K-nearest neighbor (KNN) values.
• Density is determined by the average distance to KNNs; smaller distances indicate higher density.
• Cluster centers are selected from high-density core points, with iterative merging and classification of non-core points.

Main Results:

• Experiments on manual and real-world datasets demonstrate the effectiveness of AKDPC.
• AKDPC shows higher accuracy compared to classical clustering algorithms and existing DPC variants.
• The algorithm successfully handles datasets with varying densities and non-convex structures.

Conclusions:

• AKDPC provides an effective solution for automatic cluster center selection in density-based clustering.
• The proposed method significantly improves clustering accuracy and overcomes limitations of traditional DPC algorithms.
• AKDPC offers a robust and accurate alternative for analyzing complex data structures.