An Interpretable Feature Selection Based on Particle Swarm Optimization

Liang, Yi; Wei, Qin; Zheng, Qibin; LI, Gensong; Li, Mengmeng

doi:10.1587/transinf.2021edl8095

Cited by 5 publications

(2 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Liu et al. [6] adopt different data perturbation techniques and filter feature selection methods to obtain stable feature subsets and map the subsets into initial particles. Xue et al.…”

Section: Introductionmentioning

confidence: 99%

“…For instance, Zhang et al [1] propose an improved F-measure to evaluate the quality of feature subsets, utilize fuzzy clustering to guide the initialization of the swarm, and develop a pruning strategy to eliminate the worst particles. Liu et al [6] adopt different data perturbation techniques and filter feature selection methods to obtain stable feature subsets and map the subsets into initial particles. Xue et al [7] analyse the effectiveness of different candidate solution generation strategies and propose a self-adaptive mechanism for the algorithm to adaptively select the suitable strategy to generate subsequent solutions.…”

mentioning

confidence: 99%

See 1 more Smart Citation

The classification method based on evolutionary algorithm for high‐dimensional imbalanced missing data

Liu,

Li,

et al. 2023

Electronics Letters

Self Cite

View full text Add to dashboard Cite

High dimensional imbalanced missing data classification is a challenging and complex problem that traditional algorithms struggle to solve effectively. To address this issue, a novel method is proposed, named the hybrid classification approach based on particle swarm optimization (HCPSO). HCPSO integrates ideas of feature selection, resampling and imputation, and breaks particles down into three parts. These parts represent the feature values, probabilities of resampling, and probabilities of imputation approaches, respectively. Moreover, HCPSO employs particle swarm optimization to optimize these parameters simultaneously to take advantage of these methods. Six types of algorithms, eleven datasets, and four performance indicators are used to evaluate our method. The results demonstrate a significant improvement in HCPSO's performance, with an average improvement of 13.02%, 18.95%, 20.25%, and 28.63% for accuracy, F1, AUC, and Gmean, respectively, compared to all other methods. Furthermore, the experiments also demonstrate the robustness of HCPSO.

show abstract

“…Liu et al. [6] adopt different data perturbation techniques and filter feature selection methods to obtain stable feature subsets and map the subsets into initial particles. Xue et al.…”

Section: Introductionmentioning

confidence: 99%

mentioning

confidence: 99%