All the four investigated models offer comparable classification accuracies. The choice of the one to use is, therefore, problem dependent. The classical ELM algorithm yields usually good testing accuracy while being fast to train and fast to apply to new data. The MLELM fusion of SVM's loss function and ELM's feature transformation performs poorly, because it is often unacceptably slow to train and yields accuracy similar to the classical ELM algorithm. The SVM offers state-of-the art accuracy, it is however more expensive than the ELM to both train and apply to new data. The LSSVM often matches or surpasses SVM's accuracy, and its training algorithm is simpler that the one used for SVM. However, its application to new data requires processing of the whole training set, which can be unacceptable for certain applications.