Model selection for predicting marine traffic flow in coastal waterways using deep learning methods

Accurate Marine Traffic Flow (MTF) prediction ensures the safety of ship navigation, enhances maritime services, and improves waterway navigation efficiency. This study aims to develop an innovative prediction framework that applies eight state-of-the-art Deep Learning (DL) models to predict MTF in coastal waterways, complemented by a Multi Attribute Decision Making (MADM) method that quantifies each model's accuracy, robustness, explanatory power, and computational efficiency to evaluate their applicability in various scenarios. Firstly, the Automatic Identification System (AIS) data are pre-processed to reduce the effect of missing values and outliers on the experimental results, after which the data are divided into two types: long-term and short-term. Then, the processed AIS data is applied to eight DL models for training. Secondly, expert evaluation is used to obtain five statistical indicators to monitor the performance of each method. Finally, a new model selection method based on hybrid weighted Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) is created to provide a comprehensive assessment from the perspective of accuracy and timeliness. The results indicate that Recurrent Neural Networks (RNN) is most suitable for short-term predictions, while Bidirectional Gate Recurrent Unit (Bi-GRU) is most suitable for long-term prediction needs. This study offers a baseline model for predicting MTF in coastal waterways, which is highly significant for enhancing the navigation environment, alleviating traffic pressure, improving navigation efficiency, and safeguarding navigation safety. • An integrated framework is proposed to optimize marine traffic flow prediction timeliness/accuracy. • A dual-perspective marine traffic flow prediction model is proposed to meet both short-term and long-term needs. • A hybrid weighting method is developed to select the optimal prediction model based fuzzy TOPSIS.