A multi-objective gene selection for cancer diagnosis using particle swarm optimization and mutual information

Gene expression profiling for cancer diagnosis requires the identification of optimal and non-redundant gene subsets from microarray data. We present a multi-objective particle swarm optimization (PSO) approach that balances gene-class relevancy and inter-gene redundancy by integrating mutual information. Our method employs a dual-phase search strategy: an initial PSO search followed by a local search to accelerate convergence, and a subsequent Pareto front selection to extract the non-dominated gene subsets. Experiments on cancer microarray benchmark datasets demonstrate that our approach significantly enhances feature selection and diagnosis accuracy compared to existing methods. Notably, our approach incorporates a novel dual-evaluation framework and an improved particle representation scheme, which collectively enhance robustness and prevent premature convergence. These innovations ensure a comprehensive and effective gene selection process for cancer diagnosis.