An evolutionary decomposition-based multi-objective feature selection for multi-label classification.

Area of Science:

• Data mining and machine learning
• Computational intelligence
• Information retrieval

Background:

• Multi-label classification assigns multiple categories to data instances, crucial for real-world applications.
• Feature selection is vital for enhancing multi-label classification by removing irrelevant/redundant features.
• Balancing feature reduction and classification accuracy presents conflicting objectives in multi-label feature selection.

Purpose of the Study:

• To develop a multi-objective optimization algorithm specifically for multi-label feature selection.
• To address the conflicting goals of minimizing features and maximizing classification accuracy.
• To enhance the performance and efficiency of feature selection in multi-label classification tasks.

Main Methods:

• Proposes an enhanced decomposition-based multi-objective optimization algorithm.
• Divides the multi-label feature selection problem into solvable single-objective subproblems.
• Incorporates a local search operator and a pool of genetic operators for improved solutions and diverse feature subsets.

Main Results:

• The proposed algorithm demonstrates improved performance on benchmark datasets compared to existing multi-objective feature selection methods.
• Achieved better classification accuracy with a reduced number of features.
• Evaluation metrics like hypervolume indicator and set coverage show significant improvements.

Conclusions:

• The developed multi-objective optimization algorithm effectively addresses the challenges of multi-label feature selection.
• Offers a superior balance between feature reduction and classification performance.
• Provides a promising approach for improving multi-label classification in data mining.