IMPLEMENTATION OF TRADITIONAL AND BACKPROPAGATION NEURAL NETWORK METHOD AT TRADING COMPANY TO FORECAST PRODUCT DEMAND

Abstract

Based on the data given by PT. X, there are 34,800 Kg product or material SVN stored in the warehouse which categorized as expired. The issue was coming from a bad forecast since the company did not apply standard method to execute their demand. However, after the proposed improvement was simulated by comparing 2 methods, there are improvement in the error rate.

 Backpropagation Neural Network become the best method that suitable with the pattern of SVN demand. It can be proven by the result of Mean Absolute Percentage Error (MAPE) and Mean Squared Error (MSE). The MAPE value of Backpropagation Neural Network (BPNN) with structure 2-20-1 is 17.52% with 1,613,248.51 of MSE. However, Double Moving Average Method did not show a significant improvement for forecasting the demand since the MAPE is high which yielded at 149.1% following with 18,079,648.70 of MSE.  The best forecast method that can be implemented to forecast the demand of material SVN at PT. X is by using Backpropagation Neural Network with structure 2-20-1.  By implementing the BPNN method, there is a chance to reducing the total expired stock value about 50%.

By this evidence, it is crystal clear that if the company implementing the proposed methods, the benefits would be gathered from the level of forecasting accuracy which resulting into expired stock reduction.