Enhancing Explainability in Mobility Data Science through a combination of methods

Combining multiple explanation methods works better than using just one: Rather than relying on a single explainability technique, the researchers found that using LIME, SHAP, saliency maps, attention mechanisms, and other methods together provides "deeper and more granular insights" for understanding AI models trained on movement data—suggesting that responsible AI practices benefit from layered approaches to transparency.

Standard explainability tools miss important details in trajectory data: Conventional XAI techniques "frequently overlook the distinct structure and nuances inherent within trajectory data," indicating that off-the-shelf explanation methods may not adequately serve specialized domains, and that responsible AI requires domain-specific solutions.

Different users need different types of explanations: The survey revealed "a dichotomy" where professionals with technical backgrounds (Data Scientists, ML Engineers, IT Experts) preferred more complex, amalgamated explanations, suggesting that responsible AI communication must be tailored to audience expertise rather than using one-size-fits-all approaches.

Context matters for meaningful interpretability: The framework achieves "contextually rich comprehension of model decisions" by accounting for the spatio-temporal nature of movement data, emphasizing that truly responsible AI explanation requires understanding the specific context and structure of the data being analyzed.