A practical framework for predicting residential indoor PM_2.5 concentration using land-use regression and machine...

Area of Science:

• Environmental Health
• Epidemiology
• Data Science

Background:

• Indoor air quality is critical as people spend most time indoors.
• Accurate estimation of indoor particulate matter (PM2.5) concentration is essential for epidemiological studies.
• Existing methods may not sufficiently capture the complex factors influencing indoor PM2.5 levels in residential settings.

Purpose of the Study:

• To develop and validate predictive models for indoor PM2.5 concentration in residential households.
• To identify key environmental and household factors influencing indoor PM2.5 levels.
• To assess the utility of a combined land-use regression and machine learning approach for indoor air quality assessment.

Main Methods:

• Collected data on indoor PM2.5 concentration and related parameters from 116 Hong Kong households across winter and summer.
• Estimated outdoor PM2.5 concentration using a land-use regression model.
• Applied the random forest machine learning algorithm to develop and validate indoor PM2.5 prediction models.

Main Results:

• The random forest model demonstrated high predictive accuracy (R²=0.93, cross-validation R²=0.65).
• Key predictors included outdoor PM2.5 concentration, household ID, outdoor temperature, relative humidity, household area, and air conditioning use.
• External validation confirmed the model's applicability (R²=0.47), indicating its potential for real-world exposure assessment.

Conclusions:

• A combined land-use regression and random forest approach effectively predicts indoor PM2.5 concentrations.
• The developed models offer a valuable tool for accurate exposure assessment in epidemiological research.
• This methodology enhances our ability to understand and mitigate health risks associated with indoor air pollution.