A Fast Granular-Ball-Based Density Peaks Clustering Algorithm for Large-Scale Data.

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• The Density Peaks (DP) clustering algorithm faces scalability challenges with large datasets due to its high time complexity in computing distance matrices.
• Granular Ball (GB) offers a coarse-grained data representation, grouping similar data points based on local neighborhood distributions.
• GB has been successfully applied to supervised learning tasks, enhancing efficiency for methods like Support Vector Machines and k-Nearest Neighbors.

Purpose of the Study:

• To introduce Granular Balls (GB) into unsupervised learning for the first time.
• To propose a novel GB-based DP algorithm (GB-DP) for efficient large-scale data clustering.
• To evaluate the performance of GB-DP against existing clustering algorithms.

Main Methods:

• Unsupervised partitioning generates Granular Balls (GBs) from the original data.
• Density, distance, and -distance are computed for GBs, not individual objects, eliminating parameter tuning.
• DP clustering is applied to GBs, and results are expanded to the original data.

Main Results:

• GB-DP significantly reduces running time compared to traditional DP and other algorithms like k-means and FastDPeak.
• The algorithm achieves similar or superior clustering accuracy without requiring parameter setting.
• GB-DP demonstrates effectiveness in handling large-scale datasets efficiently.

Conclusions:

• GB-DP offers a computationally efficient and effective approach for large-scale data clustering.
• The use of Granular Balls overcomes the scalability limitations of the original DP algorithm.
• GB-DP provides a parameter-free clustering solution with competitive performance.