An Evolutionary Multitasking-Based Feature Selection Method for High-Dimensional Classification.

Area of Science:

• Machine Learning
• Data Mining
• Computational Intelligence

Background:

• Feature selection (FS) is crucial for improving model performance and reducing dimensionality in data mining.
• Particle swarm optimization (PSO) is effective for FS but struggles with local optima and high-dimensional data.
• Multifactorial optimization (MFO) facilitates knowledge transfer between related tasks for complex problem-solving.

Purpose of the Study:

• To propose a novel PSO-based FS method inspired by MFO for high-dimensional classification.
• To address limitations of existing PSO-based FS methods, including local optima and computational cost.
• To leverage implicit knowledge transfer between related tasks for improved feature selection.

Main Methods:

• Developed a novel PSO-based FS method incorporating MFO principles.
• Generated two related tasks from a dataset by evaluating feature importance.
• Implemented an assortative mating crossover operator for inter-task information sharing.
• Introduced variable-range strategy and subset updating mechanism to optimize search space and population diversity.

Main Results:

• The proposed method achieved higher classification accuracy on high-dimensional problems.
• A smaller feature subset was selected compared to state-of-the-art methods.
• The method demonstrated efficiency, providing results in a reasonable time.
• Effectiveness was validated on examined high-dimensional classification datasets.

Conclusions:

• The novel PSO-MFO inspired FS method effectively addresses challenges in high-dimensional classification.
• Information sharing and introduced mechanisms enhance feature selection performance and efficiency.
• This approach offers a promising alternative to existing FS techniques for complex datasets.