Long Short-Term Memory Networks for Forecasting Demand in the Case of Automotive Manufacturing Industry

Abstract

With the rising of deep learning, neural networks have shown promising results for time series forecasting. In this paper, we investigate a deep learning-based approach for the demand forecasting method: the Long Short-Term Memory (LSTM) with the so-called Seq-2-Seq encoder-decoder architecture. To assess the performance of the proposed approach, a real-world case study was conducted for a Japanese company in the automotive manufacturing industry. In addition, the performance of the LSTM-based method is compared to the usually-used AutoRegressive Integrated Moving Average (ARIMA) method via several statistical metrics such as MSE and RMSE. The numerical experiments showed that the proposed LSTM based-approach outperforms ARIMA.