APPLYING NEURAL NETWORK FOR TIME SERIES FORECASTING FOR PRINTING COMPANY

Abstract

There are many methods in sales forecasting, but the company does not know which methods can give the best results. Company procures the raw material based on previous historical sales data estimates, without using standard methods. There is often a difference between the amount of production and the number of raw material requests. Some of the raw material is expired, which cannot be used anymore, due to less careful in planning the amount of raw material. This cause the emergence of additional costs, which make a higher inventory cost and could experience loss of sales potential. In this study, there will be an analysis and comparison between Double Exponential Smoothing (DES) – Brown, Double Moving Average (DMA), and Neural Network method. From those three-compared methods, it is found that neural network method gives more possibility to be implicated. It can be said as the best method, due to the result of error measurement that generated by Neural Network is smallest than others. Those error calculations consist of Mean Absolute Deviation (MAD), Mean Square Error (MSE), and Mean Absolute Percentage Error (MAPE). Thus, by using Neural Network will get 821.9 as the result of MAD, 1,504,894.10 for MSE and 1.79% for MAPE. The result of Validation, Verification, and Residual Test (IIDN) for the forecast result of white opaque as the raw material using neural network is pass the validation test, including verification test, tracking signal test, normality test, and the auto correlation test.