An Evolutionary Feature Selection Algorithm Using Modified Particle Swarm Optimization for High-Dimensional Genomic Datasets

International Journal of Computer Science and Technology | Vol. 12, No. 6, June 2010 | pp. 49–57

DOI: 10.46882/2010/IJCST/000092

Article Type: Original Research Paper

Title: An Evolutionary Feature Selection Algorithm Using Modified Particle Swarm Optimization for High-Dimensional Genomic Datasets

Names of Authors: Dr. Sarah Jenkins¹, Dr. Hiroshi Tanaka²

Authors’ Affiliations: ¹School of Computer Science, University of Manchester, Manchester, United Kingdom; ²Department of Bioinformatics, Kyoto University, Kyoto, Japan

Abstract: Abstract: Processing high-dimensional genomic microarrays is severely limited by the curse of dimensionality, where thousands of uninformative features degrade the classification accuracy of machine learning engines. This research presents an evolutionary feature selection algorithm driven by a Modified Particle Swarm Optimization (MPSO) metaheuristic to identify optimal subset features. The proposed MPSO variant incorporates an adaptive inertia weight mechanism and a chaotic mutation operator to prevent the particle swarm from stagnating in local optima during early evolutionary generations. The subset evaluation pipeline utilizes a support vector machine classifier acting as a wrapper fitness evaluation engine. We benchmarked the algorithm using five public cancer gene expression datasets containing up to 12,000 distinct gene dimensions. The computational outcomes show that our method reduces the necessary feature space by an average of 94.6% while increasing classification accuracy up to 98.8% ±0.4%. The average convergence duration dropped by 41.2% in comparison to generic genetic algorithms or standard binary swarm models. These results show that the adaptive mutation framework yields a highly efficient solution for high-throughput clinical diagnostic indexing pipelines.

Keywords: Feature selection, particle swarm optimization, genetic data mining, support vector machines, curse of dimensionality, evolutionary computing.

Manuscript Timeline: Received: March 05, 2010; Revised: April 11, 2010; Accepted: June 04, 2010; Published: June 16, 2010.

Citation: Jenkins, S., & Tanaka, H. (2010). An Evolutionary Feature Selection Algorithm Using Modified Particle Swarm Optimization for High-Dimensional Genomic Datasets. International Journal of Computer Science and Technology, 12(6), 49–57.

Subscribe to read the full article: https://www.internationalscholarsjournals.org/subscribe-to-read