A new model selection approach for the ELM network using metaheuristic optimization

We propose a novel approach for architecture selection and hidden neurons excitability improvement for the Extreme Learning Machine (ELM). Named Adaptive Number of Hidden Neurons Approach (ANHNA), the proposed approach relies on a new general encoding scheme of the solution vector that automatically estimates the number of hidden neurons and adjust their activation function parameters (slopes and biases). Due to its general nature, ANHNA’s encoding scheme can be used by any metaheuristic algorithm for continuous optimization. Computer experiments were carried out using Differential Evolution (DE) and Particle Swarm Optimization (PSO) metaheuristics, with promising results being achieved by the proposed method in benchmarking regression problems.