Fusing Unstructured Text and Time Series Demand and Economic Data for Demand Prediction in Air Cargo Transportation

As global trade continues to expand, air cargo transportation becomes increasingly critical. As a large portion of air cargo consists of high-value or time-sensitive products, delays and disruptions in air transportation can result in substantial economic losses. Therefore, air cargo demand prediction has become a pivotal tool to assist carriers in optimizing resource allocations and mitigating the risk of delays and disruptions. To enhance the performance of the prediction models, researchers have explored various influencing factors and data types. However, existing models in air cargo demand forecasting fail to fully integrate both quantitative and qualitative data. To address this research gap, a demand prediction approach is proposed and implemented in this study by fusing unstructured relevant textual data with time series historical demand and economic data in air cargo transportation. The study uses text embedding and a proposed multi-head CNN-LSTM model on an experimental study for John C. Munro Hamilton International Airport in Ontario, Canada. The case study involves a historical air cargo demand dataset for Hamilton Airport, a set of economic variables, and relevant social media posts on the X platform. The results show the proposed method significantly outperforms the benchmark models that do not use textual input data. Additionally, the contributions of textual inputs on prediction accuracy have been evaluated. The findings offer valuable insights for improving cargo demand management and resource allocation by improving air cargo demand forecasting through the use of multimodal inputs.