A hybrid Lévy- and differential evolution-enhanced multi-objective particle swarm optimization framework for high-dimensional cancer classification using gene selection and sample seduction

Feature redundancy, small sample size and noise in high dimensional microarray data pose serious challenge for cancer classification. To tackle these limitations, this work proposes a hybrid multi-objective optimization framework for simultaneous gene selection and sample refinement. The proposed method combines an improved particle swarm optimization strategy with fuzzy clustering to improve the relevance of features and the quality of the training data. We use a binary optimization scheme to find discriminative gene subsets and a continuous optimization scheme to select representative training samples. The two-stage strategy effective reduces the dimensionality and improves the classifier generalization. The model is evaluated using standard classification algorithms and ensemble techniques on several benchmark microarray cancer datasets. The experimental results show that the proposed framework achieves significant improvements in terms of classification accuracy, stability and computational efficiency in comparison with the conventional methods. The feature selection combined with sample optimization allows robust performance on high dimensional and imbalanced datasets, illustrating the potential of the approach for reliable cancer diagnosis and predictive analytics.