Prioritizing odorous VOCs emitted from air filters using machine learning models.

Atmospheric ozone reacts with pollutants accumulated on air filters in mechanical ventilation systems, generating odorous volatile organic compounds (VOCs). As atmospheric pollutants evolve and ozone-driven reactions intensify, new compounds may form, exacerbating odor issues. This study aims to train a machine learning framework for predicting the odor thresholds of VOCs emitted from air filters. To achieve this, machine learning models (Random Forest, Bagging Regression and Gradient Boosting) were trained based on datasets comprising 874 VOCs and 240 properties of each VOC to efficiently predict odor thresholds. Two types of used air filters were selected for a case study, with emitted VOCs were analyzed using GC-MS and HPLC at different ozone levels. Results indicated that ozone substantially increased VOC emissions from filters, with the number of detected VOC and total VOC concentrations rising by 1.1-1.6 times and 2.1-2.9 times, respectively. Random Forest model outperformed others with R = 0.786 and RMSE = 0.657. Using odor activity values, aldehydes were identified as primary odor contributors. This study identifies potential odorous VOCs on air filters, offering insights for targeted VOC monitoring and odor control.