An exploration of shipbuilding price prediction for container ships: An integrated model application of deep learning

Abstract

Shipbuilding price forecasts are key to the maritime industry's foresight, cost management, and competitive edge. This study fills a gap in the existing theoretical and empirical literature on shipbuilding price forecasting by collecting and analyzing weekly price data from October 4, 1996 to September 30, 2022, covering 17,641 observations. The study employs a CNN-BILSTM-AM model, which combines a CNN, BILSTM, AM, for shipbuilding price prediction. The findings suggest that this ensemble model effectively captures the non-linear and time-varying characteristics of shipbuilding price fluctuations. It demonstrates good adaptability to random sample selection, data frequency, and structural disruptions in the sample. This model boasts an impressive predictive accuracy, with an R ² value of 94.3 %, surpassing many standalone, composite, and traditional forecasting approaches. This study proposes a CNN-BILSTM-AM integrated model that significantly improves the shipbuilding price prediction accuracy and extends the application of machine learning in shipping economics. This study furnishes decision-support and risk management tools, utilizing historical big data to forecast shipbuilding prices, tailored for governments, financial institutions, the shipbuilding industry, and the global shipping industry.