Parcel loss prediction in last-mile delivery: deep and non-deep approaches with insights from Explainable AI

Machine learning can help e-commerce companies reduce parcel loss: The study demonstrates that predictive models can identify which parcels are at risk of being lost during delivery, enabling retailers to make better decisions about insurance and protective measures before problems occur.

Data imbalance is a real challenge in real-world applications: Since parcel loss is rare (only a very small percentage of shipments are lost), the researchers had to develop special techniques to prevent their models from simply predicting "no loss" for everything. This highlights how machine learning systems must be carefully designed to handle imbalanced real-world scenarios.

Combining different AI approaches can improve accuracy: The study found that a hybrid model combining two different machine learning techniques (a deep learning autoencoder with a random forest algorithm) outperformed simpler approaches, suggesting that thoughtfully combining multiple AI methods can yield better results than relying on a single approach.

Explainability matters for business decisions: The researchers used Explainable AI (XAI) techniques to show why their models made specific predictions about parcel loss. This transparency is crucial because it allows business managers to understand and trust the AI system's recommendations before acting on them, rather than treating the model as a "black box."

Real-world data validation strengthens AI reliability: The models were tested on a full year of actual Belgian shipment data, demonstrating that the approach works with genuine operational data rather than just theoretical datasets, which increases confidence in its practical applicability.