ABSTRACT
Machine learning performance depends on capsule network hyperparameter configuration, which the Genetic Hyper-parameter Rice Optimisation (GHPRO) method handles. Traditional hyperparameter adjustment is challenging, thus swarm intelligence and evolutionary computation algorithms were investigated. These approaches often become stuck in local optima due to random search methodologies. Inspired by China's three-line hybrid rice breeding process, GHPRO uses genetic search and hybrid rice optimisation. The hybridization and probability search methods in this unique method increase the primary HPRO. GHPRO is assessed using thirteen benchmark functions to optimise capsule networks trained on MNISTs, Chest X-Ray (COVID-19 & pneumonia) database. Experimental results show how successfully GHPRO optimises capsule network hyperparameters. By drastically boosting capsule network picture categorization performance, the technique shows its potential to improve machine learning models in numerous applications.
