# SFE: A Simple, Fast and Efficient Feature Selection Algorithm for High-Dimensional Data

**Authors:** Behrouz Ahadzadeh, Moloud Abdar, Fatemeh Safara, Abbas Khosravi, Mohammad Bagher Menhaj, Ponnuthurai Nagaratnam Suganthan  
**arXiv:** [2303.10182](https://arxiv.org/abs/2303.10182)  
**Date:** 2023-03-17  
**Summary page:** https://www.summarizepaper.com/en/paper/2303.10182v1/

## Key points

- The paper introduces a new feature selection algorithm called Simple, Fast, and Efficient (SFE) for high-dimensional datasets
- SFE algorithm operates in two phases: exploration and exploitation
- Exploration phase identifies irrelevant, redundant, trivial, and noisy features using the non-selection operator
- Exploitation phase focuses on selecting features that significantly impact classification results using the selection operator
- SFE algorithm's performance does not show significant improvement after reducing dataset dimensionality
- A hybrid approach combining SFE with Particle Swarm Optimization (PSO) is proposed as SFE-PSO to efficiently identify an optimal subset of features within the reduced search space
- Both SFE and SFE-PSO algorithms outperform existing methods significantly in feature selection from high-dimensional datasets
- Dimensionality reduction methods are crucial for improving classification accuracy by reducing noise and irrelevant information in high-dimensional datasets

## Summary

The paper presents a new feature selection algorithm called Simple, Fast, and Efficient (SFE) for high-dimensional datasets. The algorithm utilizes a search agent along with two operators - non-selection and selection - to navigate through the dataset. It operates in two distinct phases: exploration and exploitation. During the exploration phase, the non-selection operator conducts a comprehensive search across the entire problem search space to identify irrelevant, redundant, trivial, and noisy features. These features are then transitioned from selected mode to non-selected mode. In contrast, the exploitation phase involves the selection operator scouring the problem search space for features that significantly impact classification results. These influential features are switched from non-selected mode to selected mode. While the proposed SFE algorithm demonstrates success in feature selection from high-dimensional datasets, its performance does not exhibit significant improvement after reducing dataset dimensionality. To address this limitation, a hybrid approach combining SFE with Particle Swarm Optimization (PSO) is proposed as SFE-PSO to identify an optimal subset of features within the reduced search space efficiently. The effectiveness of both SFE and SFE-PSO algorithms for feature selection is evaluated on 40 high-dimensional datasets and compared against six recently developed feature selection algorithms. Results indicate that both proposed algorithms outperform existing methods significantly and can serve as efficient and effective tools for selecting features from high-dimensional datasets. such as PSO play a crucial role in enhancing machine learning algorithm efficiency by mitigating issues associated with high-dimensionality data distribution. Dimensionality reduction methods are essential in improving classification accuracy by reducing noise and irrelevant information in high-dimensional datasets. This study contributes valuable insights into  and underscores the importance of leveraging innovative approaches like  to optimize feature subset identification processes effectively. The increasing prevalence of high-dimensional datasets across various applications necessitates advanced techniques for efficient data processing and classification tasks, making the proposed SFE and SFE-PSO algorithms valuable tools in this field.