Evolving Dual-Directional Multiobjective Feature Selection for High-Dimensional Gene Expression Data

High-dimensional gene expression data has gained considerable attention in diverse medical fields such as disease diagnosis, with the challenges of the dimensionality curse and exponentially growing computation. To analyze the data, feature selection is an essential step by reducing the dimensionality. However, most feature selection algorithms for high-dimensional gene expression data still suffer from low classification and poor generalization ability. An evolutionary algorithm is an effective paradigm for enhancing global search capability in feature selection. Inspired by the evolutionary algorithm Competitive Swarm Optimization, we propose a Multiobjective Dual-directional Competitive Swarm Optimization (MODCSO) method for feature selection from high-dimensional gene expression data. First, we design a competitive swarm optimization algorithm framework based on multi-objective optimization to evolve three objective functions simultaneously. Then, we introduce a dual-directional learning strategy that trains particles within the loser group using two distinct learning strategies. To assess the effectiveness and efficiency of the suggested algorithm, we evaluate MODCSO through extensive experiments on twenty high-dimensional gene expression datasets and three real-world biological datasets. Compared to various leading feature selection algorithms, our proposed algorithm MODCSO exhibits superior competitiveness for the high-dimensional feature selection task. Moreover, we provide other extensive analyses to demonstrate further the robustness and biological interpretability of MODCSO in handling high-dimensional gene expression data.

Authors

• Yunhe Wang Huawei Noah's Ark Lab
• Zhengyu Du
• Xiaomin Li
• Wenyuan Xiao
• Hongpu Liu
• Liang Yang Hebei University of Technology

Keywords

• Feature extraction
• Optimization
• Gene expression
• Particle swarm optimization
• Linear programming
• Classification algorithms
• Bioinformatics
• Heuristic algorithms
• Convergence
• Accuracy
• High-dimensional Gene Expression Data
• Multi-objective Feature Selection
• Objective Function
• Evolutionary Algorithms
• Multi-objective Optimization
• Real-world Datasets
• Gene Expression Datasets
• Diverse Fields
• Feature Selection Methods
• High-dimensional Feature
• High-dimensional Datasets
• Feature Selection Algorithm
• Competitive Algorithm
• Global Search Capability
• Error Rate
• Classification Accuracy
• Gene Ontology Enrichment
• Dimensionality Reduction
• K-nearest Neighbor
• Average Accuracy
• Multi-objective Algorithm
• Particle Position
• Particle Velocity
• Feature Subset
• Search Direction
• Feature Selection Problem
• Filter-based Methods
• Extreme Learning Machine
• Tumor Datasets
• Pair Of Particles
• Feature selection
• competitive swarm optimization
• Algorithms
• Humans
• Gene Expression Profiling
• Computational Biology
• Databases, Genetic