Journal of Digital Information Management Chao Ma 15 3 2017 http://dline.info/fpaper/jdim/v15i3/jdimv15i3_4.pdf Recently extreme learning machine (ELM) was proposed as a new learning method for single hiddenlayer feedforward neural networks (SLFNs), it is not the same as traditional gradient based learning algorithm strategies as it can achieve good generalization performance as well as extremely fast learning speed. However, ELM may require large number of hidden neurons due to the random determination of the input weights and hidden biases, and there may exist a set of non-optimal parameters which lead ELM not be able to reach the global optimum in some cases. With the help of ideas that using a hybrid approach which takes advantage of the optimization method and ELM to train SLFNs, this study proposes a novel hybrid approach based on artificial bee colony (ABC) optimization method to optimize the ELM parameters, where the optimal input weights and biases of ELM are specified by the ABC approach and Moore- Penrose (MP) generalized inverse to analytically determine the output weights. The proposed algorithm, named ABCELM, is rigorously compared with the original ELM and other evolutionary ELM methods in different classification datasets. The obtained results clearly confirm that the proposed approach is more suitable for classification problems that we investigated, and it can not only achieve better generalization performance but be more robust with much more compact networks.