What is it about?
This article presents a comprehensive systematic review of recent advancements in optimising feature selection for cancer classification, with a specific focus on methods employing evolutionary algorithms applied to high-dimensional gene expression profiles. The review synthesises findings from relevant studies published between 2018 and 2024, retrieved from leading databases, including IEEE, Scopus, Springer, and ScienceDirect. The selected studies are categorised into four major domains: algorithm and model development for feature selection and classification (44.8%), biomarker identification using evolutionary algorithms (30%), applications of feature selection in clinical decision support systems (12%), and review or survey papers (4.5%). A central focus of this review is on the utilisation of evolutionary algorithms, such as genetic algorithms, particle swarm optimisation, and differential evolution, to address the inherent challenges of high-dimensional and low-sample-size gene expression data. Notably, this review highlights a critical research gap concerning the underexplored use of dynamic-length chromosome formulations, which offer greater flexibility and potential for enhanced optimisation performance.
Featured Image
Photo by Sangharsh Lohakare on Unsplash
Why is it important?
The significance of this review lies in its focus on addressing one of the most critical challenges in biomedical data science: the effective analysis and classification of cancer data characterised by extremely high dimensionality and limited sample sizes. Accurate classification of cancer subtypes plays a vital role in early diagnosis, prognosis evaluation, and the development of personalised treatment strategies. Feature selection serves as a critical step in reducing data complexity while preserving essential genetic markers for effective cancer classification. Evolutionary algorithms offer a robust optimisation framework for identifying the most relevant genes (features), thereby improving the predictive accuracy and computational efficiency of classification models. Despite recent advances, many existing evolutionary algorithm-based approaches still rely on static-length chromosome representations, which limit their adaptability to real-time and evolving datasets such as cancer gene expression profiles. This review highlights the potential of integrating adaptive and dynamic-length feature selection techniques within evolutionary algorithms to further enhance classification accuracy, robustness, and practical applicability in clinical settings.
Perspectives
What distinguishes this review is its forward-looking emphasis on future research and development in the field. We highlight the urgent need to move beyond conventional evolutionary algorithm-based feature selection models that rely on fixed-length features. This review advocates for the adoption of dynamic and adaptive strategies, particularly those employing variable-length chromosomes that evolve throughout the optimisation process. Additionally, this review emphasises the importance of integrating feature selection with deep learning architectures and multi-objective optimisation techniques to develop models that are not only more scalable and accurate but also computationally efficient. We further explore the transformative role of feature selection in real-world clinical decision support systems, enabling healthcare professionals to make informed and data-driven diagnostic and treatment decisions based on minimal yet highly informative gene subsets. Collectively, these perspectives call for collaborative, interdisciplinary efforts aimed at enhancing the robustness, interpretability, and clinical applicability of cancer classification systems powered by evolutionary algorithm-based machine learning.
Shir Li Wang
Universiti Pendidikan Sultan Idris
Read the Original
This page is a summary of: Feature Selection Optimisation for Cancer Classification Based on Evolutionary Algorithms: An Extensive Review, Computer Modeling in Engineering & Sciences, January 2025, Tsinghua University Press,
DOI: 10.32604/cmes.2025.062709.
You can read the full text:
Contributors
The following have contributed to this page
