Deep Learning Algorithms for Automotive Spare Parts Demand Forecasting

Forecasting the future demand of automotive spare parts precisely is important for car companies. The purchase of raw materials in advance, the plan of production, and the inventory levels are closely related to the predicted value. The accuracy of the value has an important influence on the inventory costs and customer satisfaction for car companies. Automotive spare parts demand forecasting is a typical time series forecasting problem. However, the challenge of this problem is the small amount of training data, accompanied by various kinds of spare parts and high volatility the data. In recent years, deep learning algorithms have outperformed traditional machine learning methods in many tasks and have been used to solve time series forecasting problems. Based on the real historical spare parts data of a famous automobile company in China, this paper compares the accuracy of several deep learning algorithms for the demand prediction, including fully connected networks (FCN), convolutional neural networks (CNN), long short-term memory networks (LSTM), gated recurrent unit (GRU), and transformer networks. Also, we analysis the utilities of these algorithms for the automotive sparse parts forecasting.