An Improved Extreme Learning Machine for Classification

The extreme learning machine performs better, but its performance is still influenced by hyperparameters like bias and input weights. Consequently, the Subtractive Average Based Optimizer (SABO), a classification model of the extreme learning machine, is proposed in this study to address the problems of instability brought on by the random initialization of the ELM model and the difficulty of determining the amount of nodes in the hidden layer. A SABO-based ELM model (SABO-ELM) is created by automatically optimizing the activation function and optimizing the amount of nodes in the hidden layer, input weights, and bias of the ELM. The model compares SABO-ELM's performance with three other algorithms using four benchmark datasets. Based on the experimental results, SABO-ELM outperforms all other traditional algorithms in terms of accuracy. This validates the superior performance of the SABO-ELM model by showing that the extreme learning machine optimized using the subtractive averaging optimizer performs better.