Improvement of demand forecasting system : cale study in PVC leather and plastic company

Abstract

This research presents the application of neural network to forecast the demand of the sample product. Interest rate, unemployment rate, consumer price index, oil Price, GDP, in House Garment Consumer Rate, synthetic Fiber Production, export Rate and import Rate are the input of the network which is properly train with historical sale data. The result of the forecasting is the sale volume. The learning process that we used in this thesis is backpropagation. This network is trained to be able to forecast the sale volume of sample product. For sale volume forecasting of Jul 48-Jun 49, the result from artificial neural network provides more accuracy by having the percentages of error at -1.09% with MSE at 18.78 while the result from moving average technique has the percentage of error at -5.163% with MSE at 29.165. In order to simulated the benefits of the neural network forecasting technique, the company will adjusted the production planning by using neural network forecasting instead of moving average technique. After the company adjusted the production planning according to the neural network forecasting technique, the company is successfully reducing the inventory problem. The total cost of the sample product is reducing around 2,254,000 baht which is 28%.