An improved multi-objective marine predator algorithm for gene selection in classification of cancer microarray data

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Gene selection (GS) is crucial for cancer classification, offering insights into cancer pathogenesis.
• GS in cancer classification is a multi-objective optimization problem balancing accuracy and gene subset size.
• Existing marine predator algorithms (MPA) face limitations due to random initialization and elite selection, impacting convergence and exploration.

Purpose of the Study:

• To propose a multi-objective improved MPA for enhanced gene selection in cancer classification.
• To address the limitations of random initialization and elite selection in the standard MPA.
• To improve the efficiency and effectiveness of gene selection for cancer data analysis.

Main Methods:

• A novel continuous mapping initialization using ReliefF is introduced to mitigate information loss during evolution.
• An improved elite selection mechanism employing Gaussian distribution guides population evolution towards a better Pareto front.
• An efficient mutation strategy is incorporated to prevent evolutionary stagnation.

Main Results:

• The proposed algorithm significantly reduces data dimensions in high-dimensional cancer microarray datasets.
• The enhanced MPA achieved the highest classification accuracy on most tested datasets.
• Comparative experiments against 9 other algorithms demonstrated superior performance.

Conclusions:

• The improved MPA offers a robust solution for multi-objective gene selection in cancer classification.
• The proposed initialization and selection strategies enhance the exploration and convergence capabilities of the MPA.
• This approach effectively improves cancer classification accuracy while reducing data complexity.