Application of Artificial Neural Network System for Demand Forecasting of Rubber Commodity Rss-1 Produced by an Indonesian Government-Owned Company

Abstract

Artificial neural network (ANN) system, an information computing processor, is developed to resemble the natural biological neural network system. Like the biological one, the ANN system is able to learn, store, and process or to interpret the new data input based on the previous set of values. Therefore the ANN system may be used as a tool for forecasting. This research was run to forecast demand on the rubber commodity, RSS-1, produced by an Indonesian government-owned estate company. The ANN system was modeled as a multilayer feed forward network by means of a back propagation as a learning algorithm. A sigmoid function was used as the activating function, whereas the learning rate (lr), momentum (mc), and maximum epochs 100.000 were added to accelerate the learning process. The training results showed the developed ANN system worked very well. There network operated on the structure of 2-5-4-1 with the respective value of lr = 0.01 and mc = 0.9. This structure was able to recognize the causal design between the demand, the price and stocks of RSS-1 rubber at the company by rationality of network learning results, known as MSE, at a level of 0.69%.

Keyword: Artificial neural network, demand forecasting, rubber