The clearest published example of AI in port logistics is a forecasting model that estimates how long a vessel will take to reach a port. A January 2024 study from Lamar University researchers trained an XGBoost model on Automatic Identification System (AIS) records and reported a mean absolute percentage error of about 5 percent when tested on Port of Houston data. The result concerns travel time in one set of waters, and the shared port-call data that such models depend on is a separate, regulated layer.
Key takeaways
- The study, published in the Journal of Marine Science and Engineering on January 25, 2024, used AIS records from 2018 to 2020.
- Its model, XGBoost, estimated vessel travel times with a mean absolute percentage error of about 5 percent on Port of Houston data.
- A simplified version of the model kept similar accuracy, which the authors present as a sign of lower complexity and better generalizability.
- Since January 1, 2024, IMO member states have been required to use a Maritime Single Window for port-call information.
- An arrival forecast says when a ship reaches the port, not when a berth is free.
What an arrival-time forecast predicts
An estimated time of arrival (ETA) matters to terminals, pilots, tug operators, and cargo owners, who plan work around it. Many ETAs are entered by a ship's crew and updated by hand. A machine learning ETA is instead computed from historical movement data, such as where similar ships were, how fast they moved, and how long they took to cover the same stretch of water. The task is a regression problem: given a ship's current state, predict a number of hours.
The Port of Houston study
The paper, Automatic Identification System-Based Prediction of Tanker and Cargo Estimated Time of Arrival in Narrow Waterways, is by Homayoon Arbabkhah, Atefe Sedaghat, Masood Jafari Kang, and Maryam Hamidi, all at the Department of Industrial and Systems Engineering, Lamar University. It describes a framework with two parts. A preprocessing module extracts and transforms raw AIS data and applies feature engineering. A modeling module then uses XGBoost, a tree-based machine learning method, to estimate vessel travel times.
| Element | What the paper reports |
|---|---|
| Data | Historical AIS records, 2018 to 2020 |
| Vessel types in the title | Tankers and cargo ships |
| Method | Preprocessing and feature engineering, then XGBoost |
| Test site | Port of Houston |
| Accuracy | Mean absolute percentage error of about 5 percent |
| Simplified model | Similar accuracy with lower complexity |
The abstract available for this article does not list the specific features, name other models used for comparison, or describe other waterways. Those details would need the full paper, which was not retrievable for this article.
How to read a 5 percent error
Mean absolute percentage error averages the size of each miss as a share of the true value. As an illustrative calculation, a 20-hour trip predicted with a 5 percent error would be off by about one hour on average. An average hides the spread: a model can be close on most trips and far off on a few, and the few are often the ones with bad weather or traffic that planners most want to anticipate. A percentage error also grows in absolute terms on longer voyages. The paper reports one summary figure, so a reader cannot tell from the abstract how the errors were distributed.
The data layer: Maritime Single Windows
Models like this learn from AIS, which ships broadcast, but port operations also depend on information that ships and agents report to authorities. According to the IMO page on the Maritime Single Window, amendments to the Facilitation Convention have required member states since January 1, 2024, to use a single window for the electronic exchange of information on ships' arrival, stay, and departure, with data submitted once and reused where possible. At its March 2026 session, the IMO Facilitation Committee approved draft amendments on cybersecurity for these systems, to be submitted for adoption in 2027 and expected to enter into force on January 1, 2029.
Consistent electronic records of this kind are the raw material that forecasting research needs. They also show why a forecast is only one input: IMO's separate just-in-time arrival guide draws on tracking data in which ships spent up to 9 percent of their time waiting at anchorage, which is a scheduling and contract problem as much as a prediction problem.
Reading a logistics AI claim critically
- Does the claim describe a forecast, such as arrival time, or an action, such as a changed route?
- Was accuracy measured against real arrivals in a named port and period?
- Which error measure was used, and is only an average reported?
- Was the model tested on data from after the training period or from a different port?
- Is the figure from a peer-reviewed study or from a company selling the service?
FAQ
What is AIS data?
The Automatic Identification System is a transponder system through which ships broadcast identity, position, and movement. The Houston study used historical AIS records from 2018 to 2020.
Does AI decide how fast a ship sails?
The study reviewed describes a forecasting method that estimates travel time. It does not describe a system that sets vessel speed.
How accurate are machine learning arrival forecasts?
One 2024 study reported a mean absolute percentage error of about 5 percent at the Port of Houston. Other ports, ship types, and distances may give different results.
Why does a single window matter for logistics research?
It lets authorities collect ship information once through one system. Consistent electronic data is also the raw material forecasting models need.
Related reading
- The legal framework for autonomous shipping
- AI in maritime law
- Emerging technologies and industry benefits
- AI and the future of work
More in AI Tools and Industry.
Last verified: October 2026
Sources
- Arbabkhah H, Sedaghat A, Jafari Kang M, Hamidi M, Automatic Identification System-Based Prediction of Tanker and Cargo Estimated Time of Arrival in Narrow Waterways, Journal of Marine Science and Engineering 12(2), 215, January 25, 2024
- Maritime Single Window, International Maritime Organization, 2026
- Just In Time Arrival Guide issued to support smarter, more efficient shipping, International Maritime Organization, August 11, 2020