ABSTRACT
Prediction of future values of time series is an important problem both in engineering and in science. Stochastic model building and forecasting is one of the techniques available for the analysis of discrete time series in the time-domain. Though these models have proven to be accurate for forecasting, they have several important limitations such as an a priori guess of model structure. In this paper, a feed forward neural network implementation is used to predict the future values of different time series by using past knowledge. The network is trained with statistical features such as standard deviation This hybrid neural network model is shown to be better in prediction than the traditional statistical models.
