Estimating Indoor Pollutant Loss Using Mass Balances and Unsupervised Clustering to Recognize Decays

Area of Science:

• Environmental Science
• Data Science
• Sensor Technology

Background:

• Low-cost air quality sensors are widely used indoors but often oversimplify data, losing valuable dynamic information.
• Sensor limitations include accuracy drift and lack of absolute calibration, hindering reliable data interpretation.
• Data science and machine learning offer potential solutions to enhance the utility of low-cost sensor data.

Purpose of the Study:

• To develop an unsupervised machine learning model for automated recognition of pollutant decay periods.
• To estimate indoor pollutant loss rates using time-series concentration data.
• To address limitations of low-cost sensors by extracting dynamic information and improving data analysis.

Main Methods:

• Utilized k-means and DBSCAN clustering algorithms to identify pollutant decay patterns in time-series data.
• Employed mass balance equations to calculate pollutant loss rates based on identified decay periods.
• Developed protocols for hyperparameter selection and uncertainty filtering to ensure model robustness.

Main Results:

• The model successfully identified decay periods and estimated pollutant loss rates (CO2, PM2.5) in various indoor environments.
• Observed consistently lower CO2 loss rates compared to PM2.5 loss rates within the same environments.
• Demonstrated spatial and temporal variability in pollutant loss rates, highlighting environmental influences.

Conclusions:

• The developed unsupervised machine learning model offers a novel approach to monitoring indoor pollutant removal rates.
• The model effectively overcomes limitations of low-cost sensors by analyzing temporal dynamics.
• Potential applications include evaluating ventilation and filtration systems and characterizing indoor emission sources.