Demand Forecasting Model for Cement Using Spatial Time Series Methods

Accurate demand forecasting is essential for determining optimal production and inventory management levels within a company, leading to optimized customer satisfaction levels. To obtain accurate forecasting results, it is important to select an appropriate forecasting method based on the data used. However, this is very challenging because customer demand is uncertain and fluctuates due to the influence of various external factors known as exogenous variables. Incorporating these variables into the forecasting model improves the prediction accuracy. This study aims to compare various time series forecasting methods to identify the model with the highest accuracy. The methods evaluated include Time Series Regression (TSR), Autoregressive Integrated Moving Average with exogenous variables (ARIMAX), Vector Autoregressive with exogenous variables (VARX), and the GSTAR method with exogenous variables (GSTARX). The best method is the one that produces the smallest percentage error in forecasting, as indicated by the MAPE value. This study uses cement demand data collected from a company in Indonesia, focusing on three districts in East Jawa (i.e. Jember, Malang and Surabaya). Monthly data from January 2015 to September 2023 are analyzed. The results indicate that the GSTARX method, using all three types of spatial weights, produces a smaller MAPE compared to other time series methods. Additionally, GSTARX with binary weight is the most effective method for forecasting cement demand across three locations in East Java. Therefore, based on these results, it can be concluded that adding location factors and exogenous variables to the forecasting model improves the accuracy of cement demand predictions across three locations in East Java.