Abstract:
Presents neural network techniques for forecasting the requirements of new issued banknotes. It employs Widrow-Hoff and backpropagation techniques to make the forecasts during 1993-1996 using the data in the preceding 12-15 years. It is found that the backpropagation technique provides the forecasting results closest to the actual figures with the following parameters: learning rates (10(-1) to 1), error goals (10(-2) to 10(-1)), and sumsquared error of training data (9.30x10(-4) to 3.26x10(-3)). When compared to the regression technique being used at the Bank of Thailand, this technique gives significantly more accurate results.
