Sales Forecast using a Neuro-Fuzzy Based Methodology

In the history of humanity actions around the forecast was always very important. Nowadays, large and small companies, international and national organizations, governments and educational institutions are in a continuous effort to improve the success of forecasting. Clearly, forecasts are very useful tools to avoid difficult situations and losses. On the other hand, sales as major action in a company are their main factor of their being and vital for their development. The sales forecast for them is a subject of study that is very important. This object is achieved by various methods, and one of them is the method of neural networks and fuzzy logic system. Next, as a working goal has been to present the predictability of sales by an adaptive neuro-fuzzy system (ANFIS). The ANFIS system was selected from a plurality of prediction methods, because of possibilities offered by the combination of fuzzy logic and artificial neural networks, this structure is called as ANFIS. This paper aims to apply this method on real data to prove the superiority of the results and the extraordinary possibilities of development of adaptive logic and artificial neural networks in relation to other methods, such as the method of autoregressive (AR) and the moving average regression (ARMA). In this framework are examined data, which consist of 729 observations or days of activity of a local business in the Municipality of Rethymno, Greece. During to the fact that the work objective is the percentage of results, certain errors are examined, in Anfis has better prediction, always in relation to other methods, as have been mentioned AR and ARMA. According to the results, the Afnis model had a success in predicting of 77.49%, the AR 54.8% and the ARMA 59.71%