Search

Cancer Disease remains a global health concern, demanding exploration into its causal factors for early detection and treatment. However, cancer data often presents a high-dimensional challenge for analysis. Selecting only relevant cancer genes can significantly enhance this analysis process. Traditional gene selection techniques such as heuristic methods have been employed over the years but proved infeasible. Thus, Swarm Intelligence algorithms known for their global search capabilities were developed. Nonetheless, the performance of these Swarm Intelligence algorithms is often influenced by their methods of initialization, affecting convergence, solution quality, and overall robustness. Chaos-based initialization methods have shown promise, yet their effectiveness remains underexplored in initializing SI algorithms. This research conducted a comprehensive performance comparison of three Swarm Intelligence algorithms: Particle Swarm Optimization, Salp Swarm Algorithm, and Firefly Algorithm. These algorithms were enhanced by incorporating the logistic chaotic map for initialization, specifically in the context of microarray cancer gene selection tasks. To assess the effectiveness of these enhanced algorithms, two cancer datasets were employed, namely Ovarian and Colon, and utilized two classifiers: the k-nearest neighbor and multilayer perceptron. The results of the study demonstrate that the logistic-chaos firefly algorithm paired with the k-nearest neighbor stands out as a significant performer, achieving an impressive overall accuracy rate of 93.95% while selecting 444 genes. In summary, the proposed logisticchaos firefly algorithm paired with the k-nearest neighbor approach proves itself as a worthy competitor in gene selection tasks. [ABSTRACT FROM AUTHOR]