Voting based q-generalized extreme learning machine

• We propose a novel ensemble method of extreme learning machines called V-QELM.
• V-QELM combines independent QELMs with majority voting to make the final decision.
• Simulations on 45 datasets show that V-QELM outperforms other ensemble methods.
• Kappa-error diagrams reveal that V-QELM constructs more accurate classifiers.

A novel approach to extreme learning machine (ELM) ensembles is proposed. It incorporates majority voting into the recently proposed q-generalized random neural network (QRNN) to make the final decision for classification problems. Individual ELMs are trained with q-Gaussian activation functions using different values of the parameter q (called the entropic index). As a result, these classifiers are generally more accurate than traditional ELMs. Simulations on 45 machine learning data sets show that this method, termed voting based q-generalized extreme learning machine (V-QELM), outperforms other extreme learning machine ensembles. Statistical tests (Wilcoxon, Friedman, and Nemenyi) are used to validate statistical differences between our results. Kappa-error diagrams reveal that V-QELM constructs more accurate classifiers than those found in other ensemble methods. This implies that incorporating QRNNs can lead to higher performing ensembles of extreme learning machines.