Demand Estımatıon By Sarıam Method And Support Vector Regressıon Method Perıods

Thesis Type: Postgraduate

Institution Of The Thesis: Gazi Üniversitesi, Fen Bilimleri Enstitüsü, Turkey

Approval Date: 2019




With the globalization in today's world, the competition between companies is increasing, and one step ahead in this competition depends on making strategic decisions in important activities. The production planning, sales marketing and supply chain activities come to the fore as the important and strategic decision-making activities in companies. The demand forecasting is the one of the important inputs for decision-making processes related to production planning, sales marketing and supply chain. Inaccurate demand forecasting may cause excessive product inventory, the shortage of products or loss of opportunities. Therefore, a plenty number of studies have been reported in the literature. These methods mainly estimates future values by using previous observations. The success of these methods depends on the size and complexity of the data used. In recent years, Autoregressive Integrated Moving Average (ARIMA) method and artificial intelligence-based methods (Artificial neural network, support vector regression (SVR), etc.) are frequently used in demand forecasting. In this study, the demand forecasting of four different products produced by a sweet company was performed using ARIMA and SVR methods. It was determined that seasonal effect was an important factor in the demand forecasting; and therefore Seasonal Autoregressive Moving Average (SARMA) method was used to deal with seasonal effects. To evaluate the accuracy of the proposed models, the Mean Absolute Percentage Error (MAPE) and the determination coefficient were used. When SARMA method was used, MAPE values for Product 1, Product 2, Product 3 and Product 4 were found as 19.97; 29.19; 24.58 and 21.12 respectively. MAPE values for Product 1, Product 3 and Product 4 were obtained as 11.23; 20.20 and 13.61 when RBF kernel SVR method was used, and MAPE value for Product 2 was found as 17.11 when Polynomial kernel SVR was utilized. It was observed that SVR method had higher accuracy forecasting than SARMA method when the obtained results were compared.