[Paper Review] LaDe: The First Comprehensive Last-mile Delivery Dataset from Industry

Lixia Wu, Haomin Wen+11 more|arXiv (Cornell University)|Jun 19, 2023

Urban and Freight Transport Logistics4 citations

TL;DR

This paper introduces LaDe, the first publicly available, large-scale, comprehensive, and diverse last-mile delivery dataset derived from real-world operations at Cainiao Network. It encompasses 10.7 million packages and 21,000 couriers across six months, with detailed spatio-temporal data on package locations, task events, and courier activities, enabling benchmarking for route prediction, ETA estimation, and spatio-temporal forecasting tasks with state-of-the-art baseline models.

ABSTRACT

Real-world last-mile delivery datasets are crucial for research in logistics, supply chain management, and spatio-temporal data mining. Despite a plethora of algorithms developed to date, no widely accepted, publicly available last-mile delivery dataset exists to support research in this field. In this paper, we introduce exttt{LaDe}, the first publicly available last-mile delivery dataset with millions of packages from the industry. LaDe has three unique characteristics: (1) Large-scale. It involves 10,677k packages of 21k couriers over 6 months of real-world operation. (2) Comprehensive information. It offers original package information, such as its location and time requirements, as well as task-event information, which records when and where the courier is while events such as task-accept and task-finish events happen. (3) Diversity. The dataset includes data from various scenarios, including package pick-up and delivery, and from multiple cities, each with its unique spatio-temporal patterns due to their distinct characteristics such as populations. We verify LaDe on three tasks by running several classical baseline models per task. We believe that the large-scale, comprehensive, diverse feature of LaDe can offer unparalleled opportunities to researchers in the supply chain community, data mining community, and beyond. The dataset homepage is publicly available at https://huggingface.co/datasets/Cainiao-AI/LaDe.

Motivation & Objective

• To address the lack of publicly available, large-scale, and comprehensive last-mile delivery datasets in academia and industry.
• To provide a benchmark dataset that supports research in logistics, supply chain management, and spatio-temporal data mining.
• To enable reproducible evaluation of machine learning models on real-world last-mile delivery tasks.
• To promote transparency and collaboration by releasing a dataset with detailed, anonymized, and perturbed real-world operational data.

Proposed method

• The dataset was collected from Cainiao’s real-time last-mile delivery platform, covering both package pick-up (LaDe-P) and delivery (LaDe-D) processes.
• Raw data was anonymized by obfuscating courier and package IDs, and spatial coordinates were perturbed to a 10-meter accuracy threshold for privacy.
• The dataset includes comprehensive attributes: package-level details (e.g., origin, destination, time windows), courier-level data, and task-event logs (e.g., task-accept, task-finish) with timestamps and locations.
• Data was structured into two sub-datasets: LaDe-P for pick-up and LaDe-D for delivery, each with distinct spatio-temporal patterns across multiple cities.
• Baseline models were trained and evaluated on three core tasks: route prediction, ETA prediction, and spatio-temporal graph forecasting.
• All source code and model implementations for benchmarking are publicly shared via a Hugging Face repository to support reproducibility and community extension.

Experimental results

Research questions

• RQ1 How can a large-scale, comprehensive, and diverse last-mile delivery dataset improve the evaluation and development of logistics algorithms?
• RQ2 What are the key spatio-temporal patterns and operational characteristics in real-world last-mile delivery across different urban environments?
• RQ3 To what extent can existing machine learning models generalize to real-world last-mile delivery tasks when trained on this dataset?
• RQ4 How does the inclusion of task-event logs (e.g., task-accept, task-finish) enhance the modeling of courier behavior and delivery dynamics?
• RQ5 What are the performance benchmarks for route prediction, ETA estimation, and spatio-temporal forecasting on this real-world dataset?

Key findings

• LaDe is the largest publicly available last-mile delivery dataset to date, with 10,677,000 packages and 21,000 couriers collected over six months.
• The dataset includes detailed, anonymized, and perturbed spatio-temporal data on package locations, time windows, and courier task events, ensuring privacy while preserving utility.
• Benchmarking on three tasks—route prediction, ETA prediction, and spatio-temporal graph forecasting—demonstrates the dataset’s suitability for evaluating a wide range of machine learning models.
• The dataset supports diverse research applications, including logistics optimization, route planning, and spatio-temporal forecasting, due to its comprehensive and multi-scenario design.
• The dataset is hosted on Hugging Face under a CC BY-NC 4.0 license, with source code and baseline models publicly available to encourage reproducibility and community contributions.
• The dataset is actively maintained on Hugging Face, with version updates communicated through the repository to ensure transparency and long-term usability.