Arima, multiple regression analysis and artificialneural network demand forecasting

Demand forecasting, as in many other areas, is frequently used by businesses in making various decisions such as those related to finance, logistics, and supply chain management. Especially today, considering the problem of scarce resources, establishing a balance between supply and demand is of critical importance. Due to factors such as the demands of modern life, successful marketing strategies, and population growth, the demand for dietary supplements has been increasing day by day. As a result, the number of competing companies has risen, which in turn has created a competitive environment. In today's competitive market, companies seeking to sustain their presence resort to various forecasting methods in order to manage their resources at the most optimal level. For this reason, researchers have developed various qualitative and quantitative demand forecasting methods. Demand forecasting methods are categorized into qualitative and quantitative approaches. Qualitative methods are based on experiences and intuition and are generally preferred in situations where there is a lack of numerical data regarding the item to be forecasted, where uncertainty is high, or where data variability is considerable. Quantitative methods, on the other hand, are applied in cases where data analysis is possible, utilizing historical data and statistical techniques. In this study, quantitative methods were employed based on historical sales data, specifically the Autoregressive integrated moving average (ARIMA), Multiple Regression Analysis (MRA), and Artificial Neural Networks (ANN). For the ARIMA method, historical sales quantity (D4) was used as the input variable, whereas for MRA and ANN, the input variables included historical sales quantity (D4), the exchange rate of the U.S. dollar (D1), the consumer price index for health expenditures (D2), and the retail sales volume index (seasonally and calendar-adjusted) for pharmaceutical products, medical and orthopedic products, cosmetics, and personal care items (D3). The input dataset comprised data from the years 2020, 2021, 2022, and 2023. The performance of the three models was compared based on the coefficient of determination (R2) and the Mean Absolute Percentage Error (MAPE). Three different ARIMA models were built: ARIMA (1,2,1), ARIMA (1,2,2), and ARIMA (2,2,2). The ARIMA (1,2,2) model yielded the best performance with a MAPE value of 4.9%. Multiple Regression Analysis was then conducted, resulting in a model including the adjusted D1 (D1*) and D3 as independent variables. The MRA model achieved a MAPE of 9.48%, which was lower in accuracy compared to the ARIMA result. Finally, the ANN method was applied, in which 29 out of the 48 months of data were used for network training, with the remainder used for testing and validation. The ANN model produced an error rate of 1.02%, indicating excellent performance. In conclusion, it was observed that the ANN-based forecast outperformed the other two methods in terms of accuracy.