Explanation of clustering result based on multi-objective optimization.

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Mining

Background:

• Clustering is a key unsupervised machine learning task for grouping unlabeled data.
• Traditional clustering methods lack explanatory capabilities, hindering user comprehension.
• Existing decision tree-based explanation methods often result in overly complex and difficult-to-understand outputs.

Purpose of the Study:

• To propose a novel hypercube overlay model for generating succinct and understandable explanations of clustering results.
• To address the complexity issues associated with existing explanation techniques in clustering.
• To enhance the interpretability of unsupervised machine learning outcomes.

Main Methods:

• Development of a hypercube overlay model utilizing multi-objective optimization.
• Design of two objective functions focusing on the number of hypercubes and instance compactness.
• Identification of nondominated solutions and definition of an Utopia point for optimal solution selection.

Main Results:

• The proposed model generates explanations where each cluster is covered by a minimal number of hypercubes.
• Verification on synthetic and real datasets demonstrated the model's effectiveness.
• The method provides concise and easily understandable explanations for clustering results.

Conclusions:

• The hypercube overlay model offers a significant improvement in explaining clustering results.
• The approach enhances the interpretability and usability of unsupervised machine learning.
• This method facilitates better understanding of data clusters for users.