The proliferation of high-dimensional datasets in genomics, medical imaging, and sensor networks has intensified the demand for sophisticated gene selection methodologies that can simultaneously maximize classification accuracy while achieving aggressive dimensionality reduction. Traditional metaheuristic algorithms suffer from premature convergence, inadequate boundary handling, and insufficient exploitation capabilities when applied to high-dimensional gene selection problems. This study introduces the Adaptive Intelligence Electric Eel Foraging Optimization (AIEEFO), a novel multi-strategy evolutionary framework that transforms the baseline Electric Eel Foraging Optimization through four synergistic enhancement mechanisms: (1) Chaotic Inverse Cumulative Distribution (CICD) for dynamic local refinement, (2) Predictive Trajectory-Based Motion Strategy (PTMS) leveraging historical movement patterns, (3) Periodic Mapping-Based Handling (PMBH) enabling seamless boundary traversal, and (4) Dual Memory-Driven Search Adaptation (DMSA) balancing individual and collective learning experiences. A comprehensive evaluation of the CEC2022 and CEC2017 benchmark suites demonstrated AIEEFO’s statistical superiority of AIEEFO, achieving the lowest Friedman ranks of 1.75 and 1.90, respectively, across 12 and 29 functions, significantly outperforming 12 state-of-the-art algorithms. For high-dimensional gene selection across 12 genomic datasets with up to 11,340 features, AIEEFO achieves exceptional performance: an average classification accuracy of 96.43% (ranking 1st among 13 algorithms), an average feature reduction to 112.11 features (99.05% dimensionality reduction), a superior F-score of 91.06%, and competitive fitness optimization. Statistical validation using Wilcoxon tests confirmed significant superiority over classical methods and contemporary approaches. Real-world validation was demonstrated through the ResConvNeXt+CSID+AIEEFO + D-ELM framework, developed for histopathological image classification targeting early detection of oral squamous cell carcinoma, where Convolutional Sparse Image Decomposition (CSID) fusion was employed for the deep features. Also, to address the high dimensionality of the fused feature space, the AIEEFO algorithm is applied to identify and retain the most discriminative features. The final classification is performed using a Deep Extreme Learning Machine (D-ELM), which provides efficient and accurate predictions through its multi-layer random projection and non-iterative training mechanism. The proposed framework achieves an accuracy of 97.56% and an F-score of 97.71%, outperforming traditional deep learning models and illustrating the powerful synergy between deep feature learning, intelligent fusion, evolutionary selection, and fast neural classification. This result underscores the framework’s clinical utility and the robustness of AIEEFO in high-dimensional biomedical applications.
Ibrahim et al. (Tue,) studied this question.