A Bio-Inspired Flea Optimization Framework: Performance Analysis on Benchmark Functions and Medical Diagnostic Datasets

Nature has a wide variety of survival mechanisms and social behaviors to solve very complex engineering problems for us. Meta-heuristics, inspired by various biological processes from swarm intelligence to evolutionary processes, have shown versatility in tackling non-linear, discontinuous search spaces. But the 'No Free Lunch' theorem demonstrates that there is no best optimization method applicable universally to all problems, and therefore, there is a continuous need to look for innovative methods that are inspired by nature and its ability to solve problems in very constrained domains. Metaheuristic optimization algorithms play a vital role in solving complex nonlinear and high-dimensional optimization problems. This study proposes a novel Flea Optimization Algorithm (FOA), inspired by the resilin-based jumping behavior of fleas, with the objective of achieving an effective balance between global exploration and local exploitation. The proposed FOA is tested against some of the best optimization techniques, which include Firefly Algorithm, Cuckoo Search, Genetic Algorithm, Particle Swarm Optimization, Grey Wolf Optimizer, Ant Colony Optimization, and Artificial Bee Colony. The performance of the FOA is tested using single-objective, multi-objective, and constraint functions. Moreover, three case studies are presented as application of FOA therein, FOA is used in conjunction with LASSO to optimize features, and classification is performed on the selected feature set. Besides this, an ablation study has also been conducted to investigate about the effectiveness of FOA, modified version of ResNet50 and LASSO. The experimental results show that FOA possesses competitive convergence characteristics and significantly outperforms most comparable algorithms in addressing the concerned optimization problems. In the feature optimization and classification problems, the FOA-LASSO scheme consistently obtained higher classification accuracy as compared with the other compared algorithms. For example, in Knee Joint Diagnosis, FOA provided 95% accuracy, in Colon Cancer diagnosis it achieved 99.25% accuracy and in Breast Cancer diagnosis, it reached 99.62% accuracy. Not only the accuracy, the precision, recall, F-1 score, Kappa and MCC score of proposed approach has been found promising in all cases. The results have also shown that the proposed Flea Optimization Algorithm is a reliable and flexible optimization approach that can effectively deal with a variety of benchmark problems as well as feature selection and classification problems.