A Novel Ensemble Learning Approach for Intelligent Logistics Demand Management

Abstract

Logistics demand forecasting plays a crucial role in regulating logistics management activities, developing production plans, seeking maximum economic returns, and building smart logistics. Current studies have focused on forecasting logistics demand using various statistical algorithms and machine learning models. However, it is difficult for a single learner to forecast logistics demand time series with complex nonlinear fluctuation patterns. Therefore, a novel ensemble learning approach (Deep Logistics Demand Forecasting, DeepLDF) is introduced in this work to forecast logistics demand. DeepLDF consists of two different base learners, which are the Multi-scale Time-delay Convolution Model (MSTDCM) and the Seasonal Autoregressive Integrated Moving Average (SARIMA) model. MSTDCM and SARIMA are combined for ensemble learning through a novel weight assignment approach. Based on six Singapore logistics demand data sets, DeepLDF is compared with nine different baselines. The experimental results show that DeepLDF performs well in fitting local extreme values and forecasting volatility. Overall, DeepLDF can forecast logistics demand well.