A dynamic multitask evolutionary algorithm for high-dimensional feature selection based on multi-indicator task...

Area of Science:

• Machine Learning
• Data Science
• Computational Intelligence

Background:

• High-dimensional datasets frequently contain noisy and redundant features, complicating accurate and efficient feature selection.
• Traditional feature selection methods struggle with the complexity and scale of modern datasets.

Purpose of the Study:

• To propose a dynamic multitask learning framework for robust feature selection in high-dimensional data.
• To enhance classification accuracy and reduce feature dimensionality effectively.

Main Methods:

• A dynamic multitask learning framework integrating competitive learning and knowledge transfer within an evolutionary optimization setting.
• Generation of two complementary tasks using a multi-criteria strategy for comprehensive and focused feature relevance.
• Optimization via a competitive particle swarm optimization algorithm with hierarchical elite learning and probabilistic elite-based knowledge transfer.

Main Results:

• The proposed algorithm achieved superior classification accuracy on 11 out of 13 high-dimensional benchmark datasets.
• It significantly reduced feature dimensionality, selecting fewer features than state-of-the-art methods on eight out of 13 datasets.
• Demonstrated an average accuracy of 87.24% and an average dimensionality reduction of 96.2%.

Conclusions:

• The dynamic multitask learning framework effectively balances exploration, exploitation, and knowledge sharing for robust feature selection.
• The proposed method offers a significant advancement in handling noisy and redundant features in high-dimensional data.
• Validated effectiveness through extensive experiments on benchmark datasets, showing superior performance in accuracy and dimensionality reduction.