Machine-learned modeling of PM2.5 exposures in rural Lao PDR

Hill, L. Drew; Pillarisetti, Ajay; Delapeña, Samantha; Garland, Charity; Pennise, David; Pelletreau, Aurélie; Koetting, P.; Motmans, T.; Vongnakhone, K.; Khammavong, C.; Boatman, M.R.; Balmes, John R.; Hubbard, Alan; Smith, Kirk R.

doi:10.1016/j.scitotenv.2019.04.258

Cited by 11 publications

(15 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Here, our models had R 2 values ranging from 0.23 to 0.76, indicating that overall performance was generally favorable compared that which has been previously reported. Our application of machine learning models was able to explain almost twice as much variability in exposure estimates ( R 2 0.23–0.57) than that by Hill et al 8 and the linear regression models explained up to three times ( R 2 = 0.32–0.76) that of Sanchez et al, 7 though there continues to be room for improvement in predictive power. The better performance may be the result of varying environmental factors between different study locations (eg, higher exposure contrasts), or recent improvements in measurement techniques.…”

Section: Discussionmentioning

confidence: 52%

“…Dionisio et al 52 also assessed the performance of a model in predicting child PM 2.5 exposure from personal CO, survey data, and kitchen PM 2.5 concentrations, but did not find strong relationships in any of the model permutations ( R 2 < 0.01). Hill et al 8 applied regression models and machine learning models to estimate PM 2.5 exposure ( n = 36) in a rural area of Laos, but reported adjusted R 2 values below 0.3, and RMSE of 40.0 µg/m 3 (nRMSE = 0.39). Sanchez et al 7 used stepwise regression models with survey‐based inputs to predict PM 2.5 exposures in peri‐urban South India with R 2 values ranging from 0.09 to 0.25.…”

Section: Discussionmentioning

confidence: 99%

“…Finally, we evaluated machine learning models using SuperLearner (SL) 36 as a proof‐of‐concept, following the example of Hill et al 8 SL uses cross‐validation to estimate the performance of multiple types of machine learning models or parameterizations of the same type of model. The same samples and imputation for mission data were used as for the regression modeling, though exposure estimates were was not log‐transformed as a normal distribution is not a required assumption for these machine learning models.…”

Section: Methodsmentioning

confidence: 99%

“…Currently, the available HAP exposure models, such as those used for global burden of disease estimates, have relied on simple proxies, such as primary fuel type combined with relationships to the kitchen area concentrations, 5,6 and those which have made use of survey data have shown varying degrees of predictive power 7–9 . Recent efforts by Sanchez et al 7 (peri‐urban South India) and Hill et al 8 found survey and machine learning models and survey‐based regression models were able to predict HAP exposure with R 2 values of up to 0.25 and 0.3, respectively. Shupler et al 5 modeled global HAP exposures using kitchen exposure ratios and reported R 2 values of 0.22.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Modeling approaches and performance for estimating personal exposure to household air pollution: A case study in Kenya

et al. 2021

View full text Add to dashboard Cite

This study assessed the performance of modeling approaches to estimate personal exposure in Kenyan homes where cooking fuel combustion contributes substantially to household air pollution (HAP). We measured emissions (PM2.5, black carbon, CO); household air pollution (PM2.5, CO); personal exposure (PM2.5, CO); stove use; and behavioral, socioeconomic, and household environmental characteristics (eg, ventilation and kitchen volume). We then applied various modeling approaches: a single‐zone model; indirect exposure models, which combine person‐location and area‐level measurements; and predictive statistical models, including standard linear regression and ensemble machine learning approaches based on a set of predictors such as fuel type, room volume, and others. The single‐zone model was reasonably well‐correlated with measured kitchen concentrations of PM2.5 (R2 = 0.45) and CO (R2 = 0.45), but lacked precision. The best performing regression model used a combination of survey‐based data and physical measurements (R2 = 0.76) and a root mean‐squared error of 85 µg/m3, and the survey‐only‐based regression model was able to predict PM2.5 exposures with an R2 of 0.51. Of the machine learning algorithms evaluated, extreme gradient boosting performed best, with an R2 of 0.57 and RMSE of 98 µg/m3.

show abstract

Section: Discussionmentioning

confidence: 52%

Section: Discussionmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Modeling approaches and performance for estimating personal exposure to household air pollution: A case study in Kenya

et al. 2021

View full text Add to dashboard Cite

show abstract

“…During these time periods, mothers are asked to place the vest or apron holding the equipment near their child when they are not wearing it and to leave the sampling vest where the child is expected to spend most of their time if they leave the home without their child. The vests and aprons secure the ECMs and CO loggers near the breathing zone, a similar approach to that used in other HAP exposure studies (Balakrishnan et al 2018;Bruce et al 2018;Delapena et al 2018;Hill et al 2019;Nagel et al 2016). Compliance is checked via the ECM's accelerometer data to determine if motion is detected during normal daily activities and participants are also directly asked at the end of each sampling duration about wearing the monitors as part of the survey.…”

Section: Pregnant Womenmentioning

confidence: 99%

Air Pollutant Exposure and Stove Use Assessment Methods for the Household Air Pollution Intervention Network (HAPIN) Trial

Johnson

Steenland

Piedrahita

et al. 2020

Environ Health Perspect

View full text Add to dashboard Cite

BACKGROUND: High quality personal exposure data is fundamental to understanding the health implications of household energy interventions, interpreting analyses across assigned study arms, and characterizing exposure-response relationships for household air pollution. This paper describes the exposure data collection for the Household Air Pollution Intervention Network (HAPIN), a multicountry randomized controlled trial of liquefied petroleum gas stoves and fuel among 3,200 households in India, Rwanda, Guatemala, and Peru. OBJECTIVES: The primary objectives of the exposure assessment are to estimate the exposure contrast achieved following a clean fuel intervention and to provide data for analyses of exposure-response relationships across a range of personal exposures. METHODS: Exposure measurements are being conducted over the 3-y time frame of the field study. We are measuring fine particulate matter [PM < 2:5 lm in aerodynamic diameter (PM 2:5)] with the Enhanced Children's MicroPEM™ (RTI International), carbon monoxide (CO) with the USB-EL-CO (Lascar Electronics), and black carbon with the OT21 transmissometer (Magee Scientific) in pregnant women, adult women, and children <1 year of age, primarily via multiple 24-h personal assessments (three, six, and three measurements, respectively) over the course of the 18month follow-up period using lightweight monitors. For children we are using an indirect measurement approach, combining data from area monitors and locator devices worn by the child. For a subsample (up to 10%) of the study population, we are doubling the frequency of measurements in order to estimate the accuracy of subject-specific typical exposure estimates. In addition, we are conducting ambient air monitoring to help characterize potential contributions of PM 2:5 exposure from background concentration. Stove use monitors (Geocene) are being used to assess compliance with the intervention, given that stove stacking (use of traditional stoves in addition to the intervention gas stove) may occur. CONCLUSIONS: The tools and approaches being used for HAPIN to estimate personal exposures build on previous efforts and take advantage of new technologies. In addition to providing key personal exposure data for this study, we hope the application and learnings from our exposure assessment will help inform future efforts to characterize exposure to household air pollution and for other contexts.

show abstract

A pruned feed-forward neural network (pruned-FNN) approach to measure air pollution exposure

Gong,

Liu,

Huang

et al. 2023

Environ Monit Assess

View full text Add to dashboard Cite

Environmental epidemiology studies require accurate estimation of exposure intensities to air pollution. The process from air pollutant emission to individual exposure is however complex and nonlinear, which poses signi cant modeling challenges. This study aims to develop an exposure assessment model that can strike a balance between accuracy, complexity, and usability. In this regard, neural networks offer one possible approach. This study employed a customdesigned pruned feed-forward neural network (pruned-FNN) approach to calculate the air pollution exposure index based on emission time and rates, terrain factors, meteorological conditions, and proximity measurements. The model performance was evaluated by cross validating the estimated exposure indexes with ground-based monitoring records. The pruned-FNN can predict pollution exposure indexes (PEIs) that are highly and stably correlated with the monitored air pollutant concentrations (Spearman rank correlation coe cients for 10-fold cross validation (mean ± standard deviation: 0.906 ± 0.028), for random cross validation (0.913 ± 0.024)). The predicted values are also close to the ground truth in most cases (95.5% of the predicted PEIs have relative errors smaller than 10%) when the training datasets are su ciently large and wellcovered. The pruned-FNN method can make accurate exposure estimations using a exible number of variables and less extensive data in a less money/time-consuming manner. Compared to other exposure assessment models, the pruned-FNN is an appropriate and effective approach for exposure assessment that covers a large geographic area over a long period of time.Table 1 Comparison of air pollution exposure assessment methods

show abstract

Machine-learned modeling of PM2.5 exposures in rural Lao PDR

Cited by 11 publications

References 25 publications

Modeling approaches and performance for estimating personal exposure to household air pollution: A case study in Kenya

Modeling approaches and performance for estimating personal exposure to household air pollution: A case study in Kenya

Air Pollutant Exposure and Stove Use Assessment Methods for the Household Air Pollution Intervention Network (HAPIN) Trial

A pruned feed-forward neural network (pruned-FNN) approach to measure air pollution exposure

Contact Info

Product

Resources

About