Feature selection and semi-supervised clustering using multiobjective optimization.

Area of Science:

• Computer Science
• Data Mining
• Machine Learning

Background:

• Semi-supervised clustering combines unsupervised and supervised learning to address individual limitations.
• Feature selection is crucial as not all data features are relevant for effective clustering.
• Existing methods may not optimally handle the interplay between feature selection and clustering.

Purpose of the Study:

• To develop an integrated approach for automatic feature selection and semi-supervised clustering.
• To leverage multiobjective optimization for simultaneously optimizing feature subsets and cluster parameters.
• To enhance clustering performance by identifying the most informative features.

Main Methods:

• Coupling feature selection with semi-supervised clustering using multiobjective optimization.
• Employing the Archived Multiobjective Simulated Annealing (AMOSA) algorithm for optimization.
• Encoding features and cluster centers as strings for optimization.
• Optimizing internal and external cluster validity indices along with feature count.

Main Results:

• The proposed method effectively identifies optimal feature subsets and cluster configurations.
• AMOSA successfully determined the appropriate number of clusters and data partitioning.
• Demonstrated effectiveness across seven real-life datasets with varying complexities.

Conclusions:

• The integrated semi-supervised feature selection and clustering technique offers a robust solution.
• Multiobjective optimization with AMOSA provides a powerful framework for this combined problem.
• The approach shows significant improvements over existing methods in real-world scenarios.