Predictive Modeling Of Container Throughput Using Hybrid Deep Learning And Liner Shipping Connectivity Index (LSCI) For Enhanced Port Demand Forecasting

Abstract

Predictive capacity modeling of container ports, which an integral part of the pillar of maritime transport infrastructure is planning, has the decisive influence on the effective allocation of investment. The objective of this study is to develop a hybrid deep learning algorithm based on the long-short term memory network and attention mechanism combined with the linear shipping connectivity index and its validation. Container port capacity data was gathered for 18 leading ports globally for the period from January 2010 to December 2024, and total data sample size is equal to 3,240 data points. Linear shipping connectivity index and macroeconomic variables such as GDP growth rate, containers trade volumes, marine fuel prices were incorporated into the model as key domain variables. The results indicated that the hybrid model had an average absolute error rate of 6.38% and directional accuracy of 72.85%, which meant a 46% increase compared to ARIMA (11.82%) and a 22.3% increase compared to the extended LSTM (8.21%). The importance of the variables through the permutation perturbation showed that the shipping connectivity index was the second most important variable, contributing to 19.3% in importance compared to the traditional macroeconomic variables like GDP growth, which contributed 8.1%. The contribution of the component decompositions showed that there was synergy in this case. The synergy of the attention mechanism and the connectivity index improved the results by 29.4%, while the sum of individual improvements is 26%. The decomposition analysis also showed that the benefit of the proposed model was much more evident for small ports that have more uncertainty.