Gauging-$\delta$δ: A Non-Parametric Hierarchical Clustering Algorithm.

Area of Science:

• Computer Science
• Machine Learning
• Data Mining

Background:

• Clustering is a core unsupervised learning task, crucial for data exploration and pattern discovery.
• Developing versatile, nonparametric clustering algorithms remains a significant challenge due to data complexity and unknown cluster structures.

Purpose of the Study:

• Introduce Gauging-delta, a novel nonparametric clustering algorithm designed for diverse cluster shapes.
• Address the limitations of existing algorithms in handling complex data distributions.

Main Methods:

• Employ a hierarchical merging strategy starting from individual data points.
• Utilize an adaptive mergeability function to dynamically assess cluster fusion based on perceptual statistics and environmental context.

Main Results:

• Gauging-delta demonstrates superior performance in accurately identifying well-separated clusters across 105 synthetic datasets.
• Experiments on real-world datasets underscore the algorithm's effectiveness and sensitivity to feature selection and distance metrics.

Conclusions:

• Gauging-delta offers a robust and nonparametric approach to clustering, adept at handling varied cluster geometries.
• The study highlights the importance of feature engineering and metric selection for optimizing clustering outcomes in practical applications.