Using artificial neural networks as a sales forecasting method: An application in Petkim

Along with the processing and computation power increasing parallel with the developing technology, performing complex simulations and establishing forecasting models using developed artificial intelligence technologies based on the main criterions have been rendered possible. One important application field ensuring the possibility of these models is ?Artificial Neural Networks?. Artificial Neural Networks can be defined as computer systems developed for the purpose of practicing the competencies such as producing and exploring new data by learning, which is a characteristic of human brain, automatically without any help. In this study, it is aimed to determine the method providing the highest success by comparing the forecasting performances of the ?Trend Decomposition?, ?Box-Jenkins (ARIMA) Methodology? and ?Artificial Neural Networks? which are included in the time series methods of the forecasting techniques and to forecast with the determined method the sales values of four products choosen randomly from the products being produced in Petkim for the years 1996-2006 are aimed. In this research, the sales values for the period of January 1996 ? November 2006 of four products being sold in Petkim as a value ton are utilized. In the application part of the study according to the aim of the study it is reached to conclusion that to use Artificial Neural Networks as a sales forecasting method will give more successful results and will help to make production plans in Petkim because Artificial Neural Networks having different structures relative to the given products have given lower errors compared to the other traditional time series forecasting methods.