Density Peaks Clustering by Zero-Pointed Samples of Regional Group Borders

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Density Peaks Clustering (DPC) is a popular algorithm due to its efficiency and lack of border noise.
• However, DPC lacks reliable methods for selecting the cutoff distance and automatically identifying cluster centers.

Purpose of the Study:

• To address the limitations of DPC by proposing a novel approach for automatic threshold and cluster center selection.
• Introduce Density Peaks Clustering by Zero-Pointed Samples (DPC-ZPSs) for improved clustering accuracy.

Main Methods:

• The proposed DPC-ZPSs algorithm identifies subclusters and borders using zero-pointed samples.
• Subclusters are merged based on edge sample density through an iterative process.
• This iterative merging ensures accurate determination of the cutoff distance (dc) and cluster centers.

Main Results:

• Experiments on public datasets demonstrate the effectiveness of DPC-ZPSs.
• The algorithm successfully and automatically determines the cutoff distance and cluster centers with high accuracy.
• Performance is comparable or superior to state-of-the-art clustering methods.

Conclusions:

• DPC-ZPSs offers a significant advancement over traditional DPC by providing automatic and accurate selection of key clustering parameters.
• The method enhances the reliability and applicability of density-based clustering in various data analysis scenarios.