Deep-AI soft sensor for sustainable health risk monitoring and control of fine particulate matter at sensor devoid underground spaces: A zero-shot transfer learning approach

•An AI soft sensor was developed for PM2.5 health risk predictions at metro stations.•Indoor similarity function helps for proper source-target domain adaptation.•Zero-shot transfer learning was employed to address sensor data unavailability.•Zeroshot soft sensor forecasts PM2.5 with acceptable performance without training.•Ventilation control using a transferred soft sensor improved IAQ levels by 24%.