Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
102 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Review of Swarm Intelligence-based Feature Selection Methods (2008.04103v1)

Published 7 Aug 2020 in cs.LG, cs.NE, and stat.ML

Abstract: In the past decades, the rapid growth of computer and database technologies has led to the rapid growth of large-scale datasets. On the other hand, data mining applications with high dimensional datasets that require high speed and accuracy are rapidly increasing. An important issue with these applications is the curse of dimensionality, where the number of features is much higher than the number of patterns. One of the dimensionality reduction approaches is feature selection that can increase the accuracy of the data mining task and reduce its computational complexity. The feature selection method aims at selecting a subset of features with the lowest inner similarity and highest relevancy to the target class. It reduces the dimensionality of the data by eliminating irrelevant, redundant, or noisy data. In this paper, a comparative analysis of different feature selection methods is presented, and a general categorization of these methods is performed. Moreover, in this paper, state-of-the-art swarm intelligence are studied, and the recent feature selection methods based on these algorithms are reviewed. Furthermore, the strengths and weaknesses of the different studied swarm intelligence-based feature selection methods are evaluated.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (3)
  1. Mehrdad Rostami (4 papers)
  2. Kamal Berahmand (6 papers)
  3. Saman Forouzandeh (6 papers)
Citations (293)

Summary

An Analysis of Swarm Intelligence-Based Feature Selection Methods

The paper provides an extensive review and comparative analysis of swarm intelligence-based feature selection methods, exploring their application across high-dimensional datasets. The focus is on addressing the curse of dimensionality — a prevalent issue in data mining and machine learning task — by selecting relevant, non-redundant features to enhance predictive accuracy while minimizing computational complexity.

The authors start by categorizing feature selection methods into filter, wrapper, embedded, and hybrid models, with an exploration of graph-based techniques gaining traction in recent years. The distinction between single-objective and multi-objective optimization approaches highlights the importance of coping with the dual demands of maximizing relevance to target classes and minimizing feature redundancy.

Swarm Intelligence (SI) Algorithms

The paper explores various swarm intelligence algorithms used for feature selection, such as Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), Artificial Bee Colony (ABC), Differential Evolution (DE), Gravitational Search Algorithm (GSA), and others. These algorithms, inspired by natural phenomena and biological processes, have proven effective in exploring large search spaces and finding near-optimal solutions.

Particle Swarm Optimization (PSO) emerges as a notable method due to its efficacy in maintaining a balance between exploration and exploitation, thus preventing premature convergence on suboptimal solutions. The PSO-based methods demonstrate superior performance, especially when integrating multiple classifiers for evaluation.

Ant Colony Optimization (ACO) features prominently as an effective approach, particularly in filter-based applications, owing to its robustness in uncovering relevant features by mimicking the foraging behaviors of ants.

Artificial Bee Colony (ABC) methods exploit the bee swarm's foraging patterns to optimize feature selection, often yielding promising results in terms of classifying complex datasets.

Discussions on Contributions

The paper's experimental results underscore the effectiveness of SI-based methods in significantly reducing the dimensionality of datasets while maintaining or improving classification accuracy. The PSO and ACO-based methods consistently outperform other algorithms across multiple datasets, as reflected through classification accuracy metrics and execution time benchmarks.

The paper also highlights the critical role of multi-objective optimization, which simultaneously considers multiple criteria, such as feature relevance and size, thus allowing for the derivation of more flexible and adapted feature subsets.

Implications for Future Work

While the paper explores multi-faceted SI approaches, it suggests continued research into hybrid models combining the strengths of multiple algorithms, potentially converging on an even more robust feature selection framework. Further examination of these methods in context-specific applications such as medical diagnosis or text classification remains a rich area for exploration.

The inclusion of graph-based feature selection methods points to a growing interest in leveraging data structures to enhance understanding of feature interdependencies, which could lead to more granular insights and refined selection criteria.

Concluding Thoughts

This paper offers a comprehensive evaluation of SI-based feature selection methods, emphasizing their value in reducing computational burdens and enhancing model interpretability in large-scale data mining tasks. By examining the strengths and dynamics of these algorithms, it establishes a foundational understanding for future endeavors in optimizing feature selection processes across various domains.