An improved fully informed search algorithm with population reduction and fitness-guided variable-level perturbation for medical data clustering

This paper proposes an Improved Fully Informed Search Algorithm (IFISA) for clustering medical datasets. IFISA extends the conventional FISA by integrating (i) a population reduction strategy that introduces stage-dependent adaptation into the population used to construct FISA’s fully informed mean references and (ii) a fitness-guided, variable-level perturbation mechanism that uses independently sampled, variable-specific population references and pairwise relative-fitness information to diversify search movements and help reduce the risk of premature convergence. IFISA is extensively evaluated on the CEC 2020 and CEC 2022 benchmark suites across multiple dimensions against nine metaphor-free optimization algorithms. Performance is assessed using descriptive statistics, the Friedman test, the Wilcoxon rank-sum test, boxplot visualizations, and convergence curves. An ablation study further examines the individual and combined effects of the incorporated mechanisms. To assess its practical applicability, IFISA is evaluated on 22 diverse medical datasets against the selected optimization algorithms and established clustering baselines. The results demonstrate robust and consistent performance across both benchmark suites and medical clustering datasets, with competitive or superior outcomes relative to the selected comparison methods. These findings support IFISA as a promising optimization approach for challenging global optimization problems and centroid-based medical data clustering.