A study on the deep learning based prediction of production demand by using LSTM under the state of data sparsity

Abstract

Rapid market condition changes and growing uncertainties require accurate product demand forecasting to reduce the risks of business operations. So far, recurrent neural networks (RNNs) have been introduced as the representative models for forecasting time series data. However, it showed limited performance when learning time pattern occurring over a long period of time. This study performs deep learning-based production forecasting analysis with Long-term short-term memory (LSTM) and dropout algorithm as the alternative for the conventional time series analysis methods and the exponential smoothing method. Our result provides the reference for sales target setting, facility investment and production planning, inventory control, supply chain management, and marketing strategy establishment.