In this paper, we are investigating both statistical and soft computing (e.g., neural networks) forecasting approaches. Using sales data from 1997 - 1999 to train our model, we forecasted sales for the year 2000. We found an average correlation of 90% between forecast and actual sales using statistical time series analysis, but only 70% correlation for the model based on neural networks. We are now working to convert standard input parameters into fuzzified inputs. We believe that fuzzy rules would help neural networks learn more efficiently and provide better forecasts.
展开▼
