Construction and implementation of the model for sales forecasting using neural network method

Ovaj rad se bavi istraživanjem mogućnosti upravljanja rastom i razvojem prodaje pomoću inteligentnih metoda datamining-a s naglaskom na neuronske mreže. Dan je pregled prethodnih istraživanja u tom području, te je predložen model neuronske mreže za predviđanje prodaje piva na području Vukovarsko-srijemske Županije u Hrvatskoj. Model se temelji na trinaest deskriptivnih ulaznih varijabli koje opisuju parametre pojedinog naselja, preuzetih iz Državnog zavoda za statistiku, te na podacima o prošlim prodajama piva. Vrijednost prodaje piva u sljedećem mjesecu korištena je kao izlazna varijabla. Testiran je algoritam višeslojne perceptron mreže, a dobiveni rezultati na testnom uzorku pokazuju da je model sposoban predviđati mjesečnu prodaju piva s velikom točnošću. Analiza osjetljivosti izlazne varijable pokazala je da posebno velik značaj za model imaju ovi prediktori: mjesec u godini, površine poljoprivrednog zemljišta pod voćnjacima u pojedinom naselju te broj poslovnih subjekata u promatranom naselju. S obzirom da je predviđanje budućih kretanja prodaje jedan od bitnih čimbenika u planiranju poslovnih aktivnosti, predloženi model može se koristiti kao dio integralnog sustava poslovne inteligencije poduzeća te za daljnja istraživanja u ovom području.This paper examines the possibility of managing growth and development of a company sale by using intelligent datamining methods, with the focus to neural networks. A review of previus research in this area is presented and a neural network model for predicting sale of a beer in the Vukovar-srijem County in Croatia. The model is based on thirteen descriptive input variables describing town characteristics, which were collected from the State Institute of Statistics, as well as on past sales data. The value of the total sales of beer in a town in the next month is used as the output variable. The multilayer perceptron neural network algorithm is tested, and results obtained on the test sample show that the model is able to predict monthly sale of beer with a high accuracy. The sensitivity analysis is also conducted, showing that high significance coefficients where obtained for the following predictor variables: month of the year, agricultural area planted by orchards, and the number businesses in the town. Due to the fact that the prediction of future sale movements is one of the important factors in business activity planning, the suggested model could be used as an integral part of a company business intelligence system, as well as for future research in this area