| Summary: | 碩士 === 華梵大學 === 工業工程與經營資訊學系碩士班 === 102 === The accuracy of sales forecast affects inventory level, indirectly causing the key customer satisfaction and business profitability. In this study, we introduce the neural network to do sales forecasts, and compare the performance between Elman Recurrent Neural Network (Elman) and Back-Propagation Neural Network (BPN).
This study selected input variable contains the month, monetary aggregates M1B, stock price index, the index of industrial production, manufacturing sales index, business turnover, unemployment rate, the number of employees employed regularly by industrial and service sector, monitoring indicators and other relevant information, the output variable is sales quantities. Sales data of case company ranges from January 2009 to August 2013.
In the results of Elman network, the training Mean Square Error (MSE) is 0.0067, testing MSE is 0.0026. In another result of BPN network, the training MSE is 0.0061, testing MSE is 0.0044. In order to compare the precision, the mean absolute prediction error of Elman network is 7.125 better than BPN network is 9.625, more better than the business forecast of 10.75, the empirical results can improve sales forecasting accuracy.
Keyword：Sales Forecast, Neural Networks, Recurrent Neural Network, Back-Propagation Network
|
|---|
