Optimizing Extreme Learning Machine by Animal Migration Optimization

Extreme learning machine (ELM) is a promising machine learning model for both classification and regression challenges. The ELM is very fast because it does not require training, while weights between the input layer and single hidden layer along with biases for the hidden layer are initialized randomly at the begging of execution. However, at the same time this represents a problem because for each practical classification and regression challenge, near-optimum set of weights needs to be found for satisfying performance of the model. Finding optimal (near-optimal) set of weights and biases for each practical problem is NP-hard tasks and it is known that swarm intelligence metaheuristics are able to render good results in solving these kinds of challenges. In this paper, a recently proposed animal migration optimization (AMO) algorithm is applied for tuning the ELM. Performance of this meataheuristics in this area has not been evaluated before. The ELM tuned with the AMO is validated against 4 well-known classification datasets and obtained results were compared with those generated by other machine learning models, as well as with other metaheuristics which were also used for ELM optimization. Experimental results prove that the AMO can be successfully applied in this domain.