Optimization of the backpropagation hidden layer by hybrid K-means-Greedy Algorithm for time series prediction

We propose in this paper the K-means-Greedy Algorithm (KGA) model to automate the process of finding the optimal value of the number of neurons in the hidden layer. The premise is that a backpropagation (BP) network which has this optimal number of neurons in its hidden layer would be able to produce accurate predictions of unknown values of a time series that it is trained with. We show that the proposed KGA model is effective in finding the optimal number of neurons for the hidden layer of a BP network that is used to perform prediction of a time series.