A Multi-Objective Multi-Label Feature Selection Algorithm Based on Shapley Value.

Area of Science:

• Machine Learning
• Data Science

Background:

• Multi-label learning assigns multiple labels to samples, facing challenges with high-dimensional data.
• High dimensionality can introduce noise, complicating multi-label learning.
• Feature selection is crucial for reducing dimensionality and improving model performance.

Purpose of the Study:

• To propose a Shapley value-fused feature selection algorithm for multi-label learning (SHAPFS-ML).
• To effectively identify relevant, redundant, and irrelevant features in high-dimensional datasets.
• To enhance classification accuracy and reduce computational complexity in multi-label classification tasks.

Main Methods:

• Developed a multi-objective optimization algorithm incorporating Shapley values.
• Utilized Pareto relationship to manage contradictory objectives.
• Introduced novel crossover and mutation operators based on Shapley values for feature selection.

Main Results:

• SHAPFS-ML demonstrated effectiveness in feature selection for multi-label classification.
• The algorithm successfully reduced computational complexity.
• Experimental results showed improved classification accuracy on real-world datasets.

Conclusions:

• SHAPFS-ML is an effective feature selection method for multi-label learning.
• The proposed method enhances classification performance by optimizing feature subsets.
• This approach addresses challenges posed by high-dimensional data in multi-label learning.