A Novel Clustering Method Based on Adjacent Grids Searching.

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Clustering is vital for data analysis, used in image segmentation, object recognition, and information retrieval.
• Robust clustering methods are essential due to the widespread applications of data analysis.
• Existing grid-based clustering methods face challenges with high dimensionality and cell count increase.

Purpose of the Study:

• To propose a novel clustering method, Clustering by Adjacent Grid Searching (CAGS), for robust data analysis.
• To address limitations of traditional grid clustering, such as sharp increases in cells with dimensions.
• To develop a method capable of identifying noise and cluster halos automatically.

Main Methods:

• CAGS employs a two-step strategy: adaptive grid-space construction and adjacent grid searching.
• The first step quantizes data into a multidimensional grid, distinguishing noise and halos by grid density.
• The second step uses a two-stage traversal to identify cluster cores, effectively handling arbitrary shapes and concealing halo points.

Main Results:

• CAGS successfully distinguishes noise and cluster halos based on grid density.
• The adaptive grid generation process mitigates the sharp increase in cells with data dimensions.
• Experimental results demonstrate CAGS outperforms state-of-the-art methods on various datasets, including noisy, large-scale, high-dimensional, arbitrary-shaped, varying-density, and overlapping classes.

Conclusions:

• CAGS offers a robust and effective solution for clustering diverse datasets.
• The method automatically identifies the number of clusters and handles complex data structures.
• CAGS shows significant potential for broad application in data analysis and machine learning tasks.