An inversion-based clustering approach for complex clusters.

Area of Science:

• Data Science
• Machine Learning
• Clustering Algorithms

Background:

• Traditional clustering methods often rely on feature values and assume feature independence, limiting effectiveness in real-world scenarios with inter-dependencies.
• Alternative similarity measures accounting for feature inter-dependencies are needed to improve clustering performance.

Purpose of the Study:

• To introduce the Inv measure, a novel similarity measure based on the concept of inversion.
• To evaluate the performance of a clustering approach utilizing the Inv measure on simulated data.

Main Methods:

• The Inv measure considers feature significance, object feature values, and inter-object feature values for comprehensive similarity evaluation.
• Clustering performance was assessed using the adjusted Rand index on simulated datasets.

Main Results:

• Inversion-based clustering demonstrated superior performance compared to other methods, particularly for complex and highly overlapped clusters.
• The proposed approach proves practical and effective for applications involving intricate cluster structures.

Conclusions:

• The inversion-based clustering approach shows promise in healthcare (e.g., patient identification), social media analysis (e.g., trend detection), e-commerce (e.g., customer segmentation), manufacturing (e.g., quality control), and transportation (e.g., fleet optimization).
• Its versatility makes it a valuable tool for analyzing complex, dynamic data across diverse fields.