Genetic NEAT-Based Method for Multi-Class Classification

Advances in the field of NeuroEvolution (NE) highlighted the potential of applying genetic and evolutionary mechanisms to Machine Learning (ML) problems while simultaneously alleviating the need for manual neural architecture design. The NeuroEvolution of Augmenting Topologies (NEAT) is a prominent NE method shows competitive performance in the field of reinforcement learning while its performance is not as efficient if we apply it to supervised multi-class problems. This may be attributed to the multiple objectives of the problem which hinders the learning process in NEAT. In this study, we introduce a novel method C-NEAT which is developed to address this issue and enhance the performance of NEAT without changing its core implementation. C-NEAT seeks to learn and classify different classes by creating and using a container holds the best genomes i.e. networks where each one of them focuses on learning a specific class label of the problem. Each organism in NEAT which is a unit contains the evolved genome and its fitness value is assigned automatically to a specific class based on its index in the population. During the evolution process, the container will keep updating itself and store the best evolved genomes for these classes. Through this, C-NEAT will focus on recognizing each class label and it will preserve the learning progress, thus ensuring higher learning efficiency.