Using neural nets to predict several sequential and subsequent future values from time series data

Analysing time series data in order to recognize patterns or make predictions about future values is important in many application areas, for example, failure prediction in nuclear power plants and machinery, the stock market, inventory control, marketing sales forecasts, bankruptcy prediction, forest fire predictions, etc. Solutions to many of these problems exist in nonlinear mathematics, which three layer feedforward neural nets can model. The author reports the interim results of an experimental project using a neural net approach to predict future values using time series data. Four neutral nets were constructed which predict one day in advance, based upon different sample intervals and access to economic data. Four more neural nets were constructed to predict four time periods in advance. The tests revealed that the neural net trained to sample every five days without an economic data input, outperforms the other nets