Improvement of the demand forecasting methods for vehicle parts at an international automotive company.

Abstract

This study aims to improve the forecasting accuracy for the monthly material flows of an area forwarding based inbound logistics network for an international automotive company. Due to human errors, short-term changes in material requirements or data bases desynchronization the Material Requirement Planning (MRP) cannot be directly derived from the Master Production Schedule (MPS). Therefore, the inbound logistics flows are forecast. The current research extends the forecasting methods¿ scope already applied by the company namely, Naïve, ARIMA, Neural Networks, Exponential Smoothing and Ensemble Forecast (an average of the first four methods) by allowing the implementation of three new algorithms: The Prophet Algorithm, the Vector Autoregressive (Multivariate Time Series) and Automated Simple Moving Average, and two new data cleaning methods: Automated Outlier Detection and Linear Interpolation. All the methods are structured in a software using the programming language R. The results show that as of April 2018, 80.1% of all material flows have a Mean Absolute Percentage Error (MAPE) of less than or equal to 20%, in comparison with the 58.6% of all material flows which had the same behavior in the original software in February 2018. Furthermore, the three new algorithms represent now 29% of all forecasts. All the analysis realized in this research were made with actual data from the company, and the upgraded software was approved by the logistics analysts to make all future material flow forecasts.