Forecasting seasonal time series with Functional Link Artificial Neural Network

Many economic and business time series exhibit trend and seasonal variations. In this paper, we deal with efficient modeling of time series having seasonality and definitive trends. The traditional statistical models eliminate the effect of trend and seasonality from a time series before making future forecasts. This kind of preprocessing increases the computational cost and may even degrade the forecasting accuracy. Here, we present the effectiveness of Functional Link Artificial Neural Network (FLANN) model for seasonal time series forecasting, using unprocessed raw data. The forecasting results of FLANN for four seasonal time series are compared with those of the widely popular random walk model as well as the common feedforward neural network. The comparison clearly shows that FLANN produces considerably better forecasting accuracy than all other models for each of the four seasonal time series.