Importance of the Activation Function in Extreme Learning Machine for Acid Sulfate Soil Classification

We have evaluated Extreme Learning Machine for the classification and prediction of acid sulfate soils. We have analyzed how different hyperparameters and their combinations affect the performance of Extreme Learning Machine. In addition to the activation functions, the impact of the number of hidden neurons as well as the number of features have been studied. We show that the performance of Extreme Learning Machine strongly depends on the hyperparameters. In some cases the model is unable to classify acid sulfate soils, while in others it is not only capable, but shows good accuracy in its predictions. This is largely due to the activation function. Whereas with Sigmoid or Hyperbolic tangent activation functions the model can give good results, with ReLU it does not work for classifying acid sulfate soils. In general, Extreme Learning Machine performs better for a larger number of input features. Furthermore, there is a clear correlation between the number of hidden neurons and the number of features of the input layer for the cases in which the model performs best.