Demand Forecasting of a First-Tier Supplier in Automotive Industry Using Nonlinear Autoregressive Network with Parsimonious Variables

Abstract

Accurate demand forecasting is compulsory for a first-tier supplier to determine an optimal amount of parts to produce in order to minimize safety stock after supplying to the manufacturer. Producing under an actual order will negatively impact relationships with the industry while overproducing will face unnecessary carrying costs. This study was to develop a nonlinear autoregressive exogenous network (NARX) model to predict part demands of a first-tier supplier and compare its forecasting performances with an autoregressive integrated moving average (ARIMA) model. A parsimonious set of external variables (provisional demand order and the number of non-working days) were considered in the NARX model. The time lags for each variable and demand for the previous period were determined by analyzing autocorrelation functions. The dataset was obtained from a first-tier supplier for a year and divided into 70% training, 15% validation, and 15% testing sets. The performance evaluation resulted in the root mean square error (RMSE) of the proposed model being better than an ARIMA model in both training (18%) and testing (15%) sets. The promising results of the proposed NARX model could be crucial for improving manufacturing planning to efficiently reduce carrying costs and prevent stock out.

Keywords: demand forecasting; automotive industry; neural network; parsimonious variable; ARIMA