: This paper analyzes how a robust and dynamic forecasting system was designed and implemented to predict material volumes for the inbound logistics network of an international automotive company. The system aims to reduce transportation logistics costs and improve demand capacity planning for freight forwarders. The forecasting horizon is set for 4 months and 12 months ahead in the future. To solve this problem, a time series modeling approach was carried out by using different time series forecasting methods like ARIMA, Neural Networks, Exponential Smoothing, Prophet, Automated Simple Moving Average, Multivariate Time Series, and Ensemble Forecast. Additionally, important data preprocessing methods and a robust model selection framework were used to train the models and select the best-performing one. This is known as Forward Chaining Nested Cross Validation with origin recalibration. The system performance was assessed using the Symmetric Mean Absolute Error (SMAPE). The final version of the forecasting system can deliver 4-month-ahead forecasts with a SMAPE lower than 10% for 86% of all material flow connections. The system’s forecast output is updated on a monthly basis and was integrated into the inbound logistics network system of the company.
Benzer Makaleler | Yazar | # |
---|
Makale | Yazar | # |
---|