The Adaptive Ensemble of OP-ELM Using Forward-Backward Selection
Bo Han, Bo He, Meng Meng · Applied Mechanics and Materials · 2013
Extreme learning machine (ELM) as a neural network algorithm has shown its good performance in regression or classification applications, but it has a weak robustness. In this paper, a new approach called The Adaptive Ensemble Of OP-ELM using Forward-Backward Selection (AEOP-ELM) is presented, it consists of two significant steps, firstly, we use forward-Backward selection algorithm to select the inputs which will ensure the robustness of the output, then, we train several independent OP-ELM models, and we test them iteratively to find the adaptive weights which will improve the accuracy of the output. The experiments indicate the AEOP-ELM achieves a better robustness than the original ELM as well as a better accuracy.