Q-GGXAI: a Framework for Quality-Aware Explainable AI in Geospatial Analysis

Abstract Geospatial machine learning models are widely used in domains such as urban planning, public health, and transportation. However, understanding these models is challenging due to their inherent complexity. Traditional explainable artificial intelligence methods provide either global explanations that allow broad insights into the model, or detailed local explanations for individual predictions. This study builds on an established ‘geo-glocal’ explainable artificial intelligence concept that bridges this gap by combining global and local explanations to provide a more balanced explanatory power. While the concept effectively aggregates local explanations across geospatial and temporal dimensions, it currently does not consider an essential factor: the quality of the underlying machine learning model. In this study, the existing geo-glocal concept is significantly enhanced to incorporate machine learning quality metrics into the explanation process to ensure that explanations reflect not only model predictions, but also their reliability. The concept is tested, and its versatility is demonstrated by applying it to three different real-world use cases: predicting car park occupancy, predicting rental bike bookings, and classifying accident severity. The results are visualized using an interactive matrix visualization and a novel geovisualization using multi-level glyphs.