Time Series Prediction using Backpropagation Network Optimized by Hybrid K-means-Greedy Algorithm

摘要

A multilayer perceptron with backpropagationalgorithm (BP) network that has the optimal number ofneurons in its hidden layer would be able to predict accuratelyunknown values of a time series that it is trained with. A modelknown as K-means-Greedy Algorithm (KGA) model whichcombines greedy algorithm with k-means++ clustering isproposed in this paper to find the optimal number of neuronsinside the hidden layer of the BP network. Experimentsperformed show that the proposed KGA model is effective infinding the optimal number of neurons for the hidden layer ofa BP network that is used to perform prediction of unknownvalues of the Mackey-Glass time series.