فردین اخلاقیان طاب

In this paper, a new ELM based classification method is presented to deal with the large volume of data in an efficient way. By inspiration from both parallel and sequential ELMs, this method consists of some independent ELMs which are trained using data batches in parallel. The main goal of this method is preventing exponential training time by running some ELMs in parallel which similar to the sequential methods, are trained using different data chunks. Moreover, a new aggregation method is used here to outperform this structure which can relatively achieve stable results for the different number of the ELMs. The stable results can persuade us to use such classification method on regular platforms to decrease the cost of the big data analyzing. Our method is tested on different platforms to indicate that it can be used for reducing the costs of big data analyzing. Experimental results on MNIST, KDDCup99, KDDCup99_2, Susy, and Higgs datasets shows the better performance of our method than the state-of-the-art methods