4th International Conference on Intelligent and Fuzzy Systems (INFUS), Bornova, Turkey, 19 - 21 July 2022, vol.505, pp.3-13
Extreme learning machine, a relatively recent learning algorithm that can be applied to single hidden layer feed-forward neural networks, has recently gained much attention from researchers worldwide. By virtue of its characteristics, extreme learning machine can be considered as exceptionally swift learning method with superior generalization capabilities and less required supervising by humans than other techniques. However, there are still open challenges in this domain and one of the biggest open tasks is that the capacity of extreme learning machine at large extent depends on the assigned weights and biases for the hidden layer, which represents NP-hard real-parameter optimization problem. To tackle this issue, in this research a modified variant of the animal migration optimization metaheuristics is applied for optimizing extreme learning machine hidden layer weights and biases. Suggested algorithm was tested on 7 well-recognized classification benchmarking datasets and compared with the basic animal migration optimization metaheuristics and other techniques developed thus far. According to experimental findings, proposed approach obtains improved generalization performance than the other methods.