Automating the interpretation of PM_2.5 time-resolved measurements using a data-driven approach

Area of Science:

• Environmental Science
• Data Science
• Public Health

Background:

• Automated equipment generates large volumes of time-resolved environmental data, posing interpretation challenges.
• Differentiating indoor and outdoor sources of fine particulate matter (PM2.5) is crucial for exposure assessment and health studies.

Purpose of the Study:

• To develop and evaluate an automated process for interpreting time-resolved PM2.5 data.
• To distinguish between indoor and outdoor PM2.5 emission sources using machine learning.

Main Methods:

• Utilized Random Forest (RF), a machine learning algorithm, to analyze PM2.5 data.
• Trained the model on a dataset of 836 indoor emission events from 18 California apartments over two weeks.
• Evaluated model performance based on sample size, source variation, and data characteristics.

Main Results:

• Data from numerous events across different locations are essential for robust model generalizability.
• Longitudinal data proved more effective for source identification than high-frequency measurements within a single location.
• The developed RF model successfully identified 442 indoor emission events in a separate dataset of 65 homes with minimal misidentifications.

Conclusions:

• An automated RF-based approach can effectively interpret time-resolved PM2.5 data and identify indoor emission sources.
• The study highlights the importance of diverse and longitudinal data for building accurate environmental exposure models.
• This method offers a scalable solution for analyzing large datasets in indoor air quality research.