Boosting Algorithms for Delivery Time Prediction in Transportation Logistics

TL;DR

This paper evaluates various machine learning algorithms, especially boosting methods, for predicting delivery times in postal services, demonstrating their effectiveness in reducing delays and outperforming traditional models.

Contribution

It demonstrates the applicability of advanced boosting algorithms like LightGBM and CatBoost for long-term delivery time prediction in postal logistics, with extensive experimental validation.

Findings

Boosting algorithms outperform linear regression and random forest in accuracy.

LightGBM and CatBoost show higher efficiency and better performance.

Travel time prediction helps reduce delays in postal services.

Abstract

Travel time is a crucial measure in transportation. Accurate travel time prediction is also fundamental for operation and advanced information systems. A variety of solutions exist for short-term travel time predictions such as solutions that utilize real-time GPS data and optimization methods to track the path of a vehicle. However, reliable long-term predictions remain challenging. We show in this paper the applicability and usefulness of travel time i.e. delivery time prediction for postal services. We investigate several methods such as linear regression models and tree based ensembles such as random forest, bagging, and boosting, that allow to predict delivery time by conducting extensive experiments and considering many usability scenarios. Results reveal that travel time prediction can help mitigate high delays in postal services. We show that some boosting algorithms, such as light gradient boosting and catboost, have a higher performance in terms of accuracy and runtime efficiency than other baselines such as linear regression models, bagging regressor and random forest.