Genetic Algorithm and the Kruskal–Wallis H-Test-Based Trainer Selection Federated Learning for IoT Security

Federated learning of a decentralized machine learning approach is used for attack detection which trains models collaboratively across multiple IoT devices. The dynamic selection of training nodes is essential due to the heterogeneity of IoT devices. This research presents a new framework for trainer selection in federated learning for IoT security using genetic algorithms and the Kruskal-Wallis H-test. A genetic algorithm is used for the optimal selection of trainers based on computational capabilities, bandwidth, and security. The Kruskal-Wallis H-test, a non-parametric statistical test is used as the objective function to ensure the selected trainers have statistically significant diversity. This combined approach outperforms random and fixed trainer selection methods and improves model accuracy, robustness, and security.