Enhancing the performance of hybrid evolutionary algorithms in microarray colon data classification

Abstract

Microarray technology aids in monitoring simultaneous gene expressions of human cells for diagnosis and treatment of infectious diseases in a single experiment. The microarray data are of high dimension with a large number of genes and fewer samples. Highly informative features and appropriate classifiers are necessary to increase the accuracy of disease classification. In this paper, a publicly available colon cancer dataset is analyzed for classification. At first, Power Spectral Density (PSD) technique is employed to obtain the reduced gene features and after that, they are classified using various types of classifiers for colon cancer classification. To improve the classification accuracy, the hybridized evolutionary classifiers, namely, Artificial Bee Colony‐Firefly (ABC‐Firefly), Artificial Bee Colony‐Particle Swarm Optimization (ABC‐PSO), Particle Swarm Optimization‐Firefly (PSO‐Firefly) are derived from the conventional evolutionary classifiers, namely, Artificial Bee Colony (ABC), Particle Swarm Optimization (PSO) and Firefly classifiers. The performance of the three hybridized evolutionary algorithms is compared with the conventional classifiers. Additionally, the comparison was also performed on the conventional and hybridized evolutionary algorithms using Expectation Maximization (EM) and cascaded Power Spectral Density‐Expectation Maximization (PSD‐EM) dimensionality reduction techniques. Although there are many classifiers for colon classification, it is wise to use PSD features with a Hybrid ABC‐PSO classifier for enhanced performance. The experimental results reveal that the Hybrid ABC‐PSO classifier with PSD features achieves 98.96% for colon cancer samples and 100% accuracy for colon normal samples when compared with the other classifiers reported in the literature.