An Explainable Stacked Ensemble Model for Static Route-Free Estimation of Time of Arrival

Ensemble models improve taxi arrival time predictions but create a "black box" problem: The paper demonstrates that combining multiple machine learning models into a stacked ensemble increases prediction accuracy for estimated time of arrival (ETA), but this sophistication makes it harder to understand how the system reaches its conclusions—a critical issue when passengers and drivers rely on these predictions for planning.

Explainable AI can make complex prediction systems transparent: By applying explainability techniques to both layers of their ensemble model, the researchers show that it's possible to "open the black box" and identify which factors (such as traffic conditions, time of day, or distance) actually drive the ETA predictions, rather than treating the system as an inscrutable algorithm.

The model correctly learned real-world factors that matter for trip duration: The experimental evaluation confirmed that the system appropriately weighted important features—suggesting the model isn't relying on spurious correlations or biases, but rather on factors that genuinely affect how long a taxi trip takes, which is essential for trustworthy AI systems.

Explainability methods can be adapted for complex AI architectures: The paper proposes new techniques for combining explanations across multiple model layers, demonstrating that transparency doesn't have to be sacrificed when using sophisticated AI approaches—a key consideration for deploying responsible AI in real-world applications.