Nonlinear time series prediction using artificial neural network

Time series is a sequence of data which is obtained from measurements of time dependent phenomenon. Time series includes data points in the past and it can be measured from various phenomenons in real world. The procedure for determining the future values of the series, by using the past values, is called as time series prediction. Various statistical methods can be used for time series prediction or a linear or nonlinear mathematical model can be suited to time series. Time series may include some economic, social, seasonal and speculative fluctuations. Furthermore, time series can also be governed by nonlinear and/or chaotic dynamics. In this case, time series prediction using a suitable mathematical model may be necessary, instead of statistical methods.In this thesis, time series prediction using artificial neural network is examined. Neural networks are also widely used in signal prediction as they can be used in various fields, due to their nonlinear mathematical model. In this study, gold prices having a regular trend and Mackey-Glass chaotic time series are used for testing of neural network predictor. Short and mid term prediction of these time series are implemented using feed-forward neural network. Various neural network structures using different data points as inputs are tested and prediction performances are determined.