3 Jan 2024 | Yinglan Feng, Liang Feng, Senior Member, IEEE, Songbai Liu, Member, IEEE, Sam Kwong, Fellow, IEEE, and Kay Chen Tan, Fellow, IEEE
The paper introduces a novel evolutionary multitasking (EMT) framework, MO-FSEMT, for multi-objective high-dimensional feature selection (FS). Traditional FS methods often suffer from limitations such as inadequate knowledge acquisition, exploitation, and transfer, leading to suboptimal solutions. MO-FSEMT addresses these issues by constructing multiple auxiliary tasks using distinct formulation methods, providing diverse search spaces and information representations. Each task is optimized independently using separate evolutionary solvers with task-specific representations and biases, ensuring efficient and effective knowledge transfer. The proposed framework includes a multi-manner-based problem formulation strategy, a multi-solver-based multitask optimization scheme, and a task-specific knowledge transfer mechanism. Experimental results on 26 high-dimensional datasets demonstrate that MO-FSEMT outperforms state-of-the-art FS methods in terms of classification accuracy and computational efficiency. Ablation studies validate the effectiveness of each component in the MO-FSEMT framework.The paper introduces a novel evolutionary multitasking (EMT) framework, MO-FSEMT, for multi-objective high-dimensional feature selection (FS). Traditional FS methods often suffer from limitations such as inadequate knowledge acquisition, exploitation, and transfer, leading to suboptimal solutions. MO-FSEMT addresses these issues by constructing multiple auxiliary tasks using distinct formulation methods, providing diverse search spaces and information representations. Each task is optimized independently using separate evolutionary solvers with task-specific representations and biases, ensuring efficient and effective knowledge transfer. The proposed framework includes a multi-manner-based problem formulation strategy, a multi-solver-based multitask optimization scheme, and a task-specific knowledge transfer mechanism. Experimental results on 26 high-dimensional datasets demonstrate that MO-FSEMT outperforms state-of-the-art FS methods in terms of classification accuracy and computational efficiency. Ablation studies validate the effectiveness of each component in the MO-FSEMT framework.