A decision tree model for the prediction of the stay time of ships in Brazilian ports. (January 2023)

• Record Type: Journal Article
• Authors: Abreu, Levi R.; Maciel, Ingrid S.F.; Alves, Joab S.; Braga, Lucas C.; Pontes, Heráclito L.J.
• Is Part Of: Engineering applications of artificial intelligence. Volume 117:Part B(2023)
• Publication Date: 2023-01
• Subjects: Waterway modal -- Brazilian ports -- Data mining -- Stay time -- Random forests; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2022.105634
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: Maritime transport is an alternative modal logistic in transporting cargo for long distances and in large quantities. However, the logistical planning for this modal becomes costly due to the uncertainties, such as climatic conditions, cargo types, and port characteristics. Therefore, estimating the stay times of ships becomes an essential objective for the planning and scheduling of the waterway modal. Determining the time frame the port has to operate the ship, based on the expected time that ships stay moored, is a challenge for the port management. In the present study, we collected data on the main cargo movements in Brazilian ports in 2018 to develop a model for predicting the stay time of ships, using algorithms based on decision tree models. There are no studies in the literature on models for predicting the stay time of ships, which is the gap to be filled in this research. In addition, an exploratory data analysis was performed to discover the variables that most influence the stay time. This research used several classification machine learning algorithms (support vector machines, gradient boosting, decision tree, random forest, among others) to build stay time prediction. As a result, the best model generated was that of random forests that obtained acceptable performance, with accuracy and f1-score above 73%, and train and test times around 14 s and the most important features to the model involve geographical and cargo characteristics. Therefore, it is possible to use them in real environments to develop logistic planning of the waterway modal.