Orca Predator Algorithm for Feature Selection

In the era of data explosion, the volume and dimensionality of information pose significant challenges to the accuracy and effectiveness of machine learning systems. An efficient alternative to address this challenge is the feature selection problem, which aims to find a subset of components that provide similar results with less computational effort. Feature selection is known as an NP-hard combinatorial problem due to the exponential growth in the number of possible feature subsets as the total number of features increases. In this study, we propose a mechanism for the feature selection using the Orca Predation Algorithm, a novel metaheuristic inspired by the hunting behavior of orcas. This approach has been underexplored for solving combinatorial problems and has shown excellent results in recent applications. We evaluate the performance of the metaheuristic on an electrocardiogram dataset obtained from Kaggle using five machine-learning classification algorithms. Our results indicate that applying the bio-inspired algorithm for feature selection can enhance the performance of these algorithms. For evaluation, we employ three key metrics: F1 score, accuracy, and density. The obtained results show that 4 of 5 hybridizations exhibit improvements. This study opens up new possibilities for the use of this metaheuristic in other problems of similar complexity.