Predicting Last-Mile Delivery Route Deviations Using Machine Learning
Published in Expert Systems with Applications, 2025
This paper presents an end-to-end machine learning pipeline for predicting delivery route deviations in last-mile logistics. The study analyzes over 20,000 historical routes and evaluates multiple neural architectures including LSTM, CNN, and Transformer models.
Key Contributions
- Comprehensive analysis of 20,000+ historical delivery routes
- Evaluation of three neural architecture approaches (LSTM, CNN, Transformers)
- Achieved 19% improvement in prediction accuracy over existing baseline methods
- Practical application for Norwegian logistics company operations
Results
The proposed ML pipeline successfully identifies factors affecting route deviations with significant accuracy improvements compared to traditional prediction methods.
Recommended citation: Konovalenko, A., Hvattum, L.M., & Iversen, K.A.H. (2025). "Predicting Last-Mile Delivery Route Deviations Using Machine Learning." Expert Systems with Applications. https://doi.org/10.1016/j.eswa.2025.129921
Download Paper