A Comparison of Forecasting Building Material Inventory between Backpropagation Neural Network And Arima

11th International Seminar on Industrial Engineering & Management, "Technology and Innovation Challenges Towards Industry 4.0 Era

工业技术（总论）

Soenandi, I.A.^1 ; Hayat, C.^2

Department of Industrial Engineering, Faculty of Engineering and Computer Science, Krida Wacana Christian University, Tanjung Duren Raya No. 4, Jakarta

11470, Indonesia^1

Department of Information System, Faculty of Engineering and Computer Science, Krida Wacana Christian University, Tanjung Duren Raya No. 4, Jakarta

11470, Indonesia^2

关键词: Arima;    Back propagation neural networks;    Forecasting modeling;    inventory;    Material inventories;    Performance criterion;    Tentative models;    Time series modeling;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/528/1/012044/pdf
DOI  :  10.1088/1757-899X/528/1/012044

学科分类：工业工程学

来源: IOP

【 摘 要 】

The demand for forecasting task is very important to determine the number of stocks efficiently. This process should accommodate the demand for a company's product or service and control the inventory level. Especially for products such as building materials that needed capitals to buy and wide space to keep it safe. This research has objective to minimize the excessive amount of product in inventory and minimize loss in sales. This study was compared between a method named Back Propagation Neural Network (BPNN) that known as one of the most accurate and widely used forecasting model and ARIMA as a time series model to find the most accurate in forecasting of inventory. In this case, the model of BPNN used 6 input neurons as a monthly period of sale, the price of the product, amount of historical selling, an approximation of project renovation, an approximation of new project building and number of a competitor. And for Arima method we have three trials of tentative models. To compare the accuracy between them, we used the performance criteria such as MAD, MAE, RMSE, RRSE and RAE. In this research, we obtained that forecasting with BPNN is more accurate than ARIMA with error prediction of 19.6, 19.6, 30.4, 0.6, 0.5 for those performance criteria consecutively.