Novel density-based and hierarchical density-based clustering algorithms for uncertain data.

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Traditional clustering algorithms struggle with uncertain data.
• Existing density-based methods like FDBSCAN and FOPTICS have limitations, including information loss, high complexity, and nonadaptive thresholds.

Purpose of the Study:

• To address limitations in existing density-based clustering algorithms for uncertain data.
• To introduce novel algorithms that improve accuracy, efficiency, and handling of data uncertainty and varying densities.

Main Methods:

• Proposed PDBSCAN algorithm with improved probability computation and new definitions for probability neighborhood, support degree, core object probability, and direct reachability probability.
• Developed PDBSCANi, an enhanced version of PDBSCAN, with a refined cluster assignment strategy.
• Introduced POPTICS, a hierarchical density-based algorithm, extending PDBSCAN with fuzzy core and reachability distances for non-uniform densities.

Main Results:

• PDBSCAN and PDBSCANi demonstrated reduced complexity and solved nonadaptive threshold issues present in FDBSCAN.
• POPTICS effectively revealed cluster structures in datasets with varying local densities, surpassing PDBSCAN and PDBSCANi.
• Experimental results confirmed the proposed algorithms' superiority in accuracy and efficiency over existing methods.

Conclusions:

• The novel PDBSCAN, PDBSCANi, and POPTICS algorithms offer significant improvements for clustering uncertain data.
• These algorithms effectively handle data uncertainty, varying densities, and computational complexity, advancing density-based clustering techniques.