1777. Integrating Probabilistic ETA Forecasting to Enhance On-Time Delivery in Last-Mile Logistics
Invited abstract in session WC-56: Logistics, stream Vehicle Routing and Logistics.
Wednesday, 12:30-14:00Room: Liberty 1.11
Authors (first author is the speaker)
| 1. | Marie-Christine Heller
|
| Logistics Management, WHU - Otto Beisheim School of Management |
Abstract
Last-mile delivery remains one of the most cost-intensive and unpredictable stages in logistics, with growing pressures from increasing parcel volumes, stricter city regulations, and evolving consumer expectations. Traditional scheduling methods rely on static ETA calculations that fail to adapt to real-time disruptions, leading to frequent delivery delays and inefficiencies.
This study introduces a hybrid predictive modelling framework that shifts from deterministic to probabilistic ETA forecasting, enabling dynamic time-window adjustments to improve On-Time Delivery (OTD) reliability. Instead of a single ETA estimate, the model predicts ETA distributions using quantile regression (5%, 50%, 90%) to capture uncertainty.
The approach integrates:
• Historical and real-time data (traffic, weather, driver performance)
• Multi-task learning models, e.g. Gradient Boosting Machines (GBMs) and Long Short-Term Memory (LSTM) networks to optimize ETA predictions and confidence intervals
• Cost functions that penalize deviations outside the probabilistic range
• Benchmarking against static scheduling methods using Mean Absolute Error (MAE) and delivery adherence metrics
Predictive analytics enables a shift from reactive to proactive customer service. The proposed framework not only improves OTD reliability but also contributes to optimizing resource allocation and enhancing customer satisfaction.
Keywords
- Logistics
- Analytics and Data Science
- Vehicle Routing
Status: accepted
Back to the list of papers