An extended affinity propagation clustering method based on different data density types.

Area of Science:

• Data Science
• Machine Learning
• Algorithm Development

Background:

• Traditional Affinity Propagation (AP) clustering requires homogeneous data distributions.
• Non-homogeneous datasets with varying data intensity pose challenges for standard AP clustering.
• Existing methods may struggle to identify distinct clusters in complex data.

Purpose of the Study:

• To propose an extended Affinity Propagation (AP) clustering algorithm for non-homogeneous datasets.
• To improve the accuracy of cluster grouping in data with varying densities.
• To provide a more robust clustering solution compared to standard AP and OPTICS.

Main Methods:

• Data points are initially partitioned into density types based on nearest neighbor distances.
• The Affinity Propagation (AP) algorithm is then applied independently within each density type.
• The performance is evaluated using both artificial and real-world seismic datasets.

Main Results:

• The extended AP algorithm demonstrated superior accuracy in grouping data points compared to standard AP.
• The proposed method outperformed the OPTICS algorithm in clustering performance on tested datasets.
• Accurate cluster identification was achieved even in datasets with heterogeneous data distributions.

Conclusions:

• The enhanced AP clustering algorithm effectively addresses limitations of standard AP in non-homogeneous data.
• This novel approach offers improved accuracy and robustness for clustering complex datasets.
• The method shows significant potential for applications in data analysis and pattern recognition.