Fewer Unproductive Moves at a Container Terminal
Most yard moves at a container terminal earn nothing: containers are dug out, re-stacked, and shuffled aside to reach others. At the terminal in the study, three of every four movements were unproductive, and about half of those traced to pre-clearance service requests — the administrative previo a customs broker asks for, distinct from formal customs inspection — that arrive with little warning (Villalobos et al. 2026).
The terminal's operating rules flagged which containers would request pre-clearance service with 25 % precision and 30 % recall — most service requests arrived as surprises, and every surprise meant digging a container out of a stack built without it in mind.
A real terminal's complete operational history, with consignees deduplicated across naming variants and cargo described by its HS classification. All figures below come from the published preprint, evaluated with temporal validation — each model tested only on months it never saw.
Two supervised models: which containers will request pre-clearance service, and how long each container will dwell in the yard. Features draw on consignee history and cargo classification rather than weather-and-seasonality generalities.
Service prediction reached 75 % precision and up to 80 % recall on the top-300 ranking, against 25 % and 30 % for the rules in force; the preprint estimates that placement informed by these models avoids 13 % to 23 % of total handling operations (arXiv:2604.06251, §7.5). The next step — the optimizer that turns predictions into yard positions and gate appointments — is what the models exist for.