| || || Learning | Mathematical models|
| || || Cooperative neuro-evolution of recurrent neural networks for multiple step time series prediction |
Author: Hussein, Shamina
Institution: The University of the South Pacific
Subject: Evolutionary computation, Neural networks (Computer science), Learning | Mathematical models
Call No.: Pac TA 347 .E96 H87 2017
Copyright:This thesis may NOT be copied without the authors written permission.
Abstract: Multi-step ahead prediction has been one of the greatest challenges for machine learning. Neuro-evolution is the use of evolutionary algorithms to train neural networks that have been used for time series prediction. Recurrent neural networks are one of the major neural network architectures that have the ability to efficiently model temporal sequences. They have been trained using cooperative neuro-evolution and backpropagation through-time. In this thesis, a statistical feature selection method is presented that employs varying feature windows for time series prediction.