Explainable AI-driven high-fidelity IAQ prediction (HiFi-IAQ) model for subway stations: Spatiotemporal outdoor air quality interpolation using geographic data

Indoor air quality (IAQ) in underground subway stations is an emerging concern regarding the health of passengers. Outdoor air quality (OAQ) is strongly correlated with IAQ; thus, indoor air should be considered comprehensively with outdoor air. However, lack of OAQ monitoring stations resulted no outdoor measurements at various subway stations. To tackle this issue, the high-fidelity IAQ prediction (HiFi-IAQ) model was developed with spatiotemporal OAQ interpolation using geographic information system (GIS). First, a city-wide OAQ measurements were collected, and a graph convolutional network (GCN) model was developed to interpolate spatiotemporal OAQ at a target subway station. Based on the interpolated GIS-driven OAQ (InterGIS-OAQ), the HiFi-IAQ model was developed and explainable artificial intelligence (XAI) was used for the interpretation and improvement of the fidelity of a model. The results reveal that InterGIS-OAQ prediction at the target subway station can capture the spatiotemporal dynamics of OAQ from the citywide 23 OAQ monitoring stations and the HiFi-IAQ model with InterGIS-OAQ exhibits outstanding predictive performance with R², mean absolute error, and root mean squared error of 0.89, 3.02, and 5.96, respectively. Hence, the HiFi-IAQ offers a high-fidelity and explainable prediction and guides operator toward efficient ventilation in subway stations.