Please use this identifier to cite or link to this item:
http://ir.mu.ac.ke:8080/jspui/handle/123456789/9355
Full metadata record
DC Field | Value | Language |
---|---|---|
dc.contributor.author | Mangeni, Judith | - |
dc.contributor.author | Upadhya, Adithi Raghavendra | - |
dc.date.accessioned | 2024-08-21T10:20:28Z | - |
dc.date.available | 2024-08-21T10:20:28Z | - |
dc.date.issued | 2024-08 | - |
dc.identifier.uri | http://ir.mu.ac.ke:8080/jspui/handle/123456789/9355 | - |
dc.description.abstract | BACKGROUND AND AIM[|]Household air pollution (HAP) exposure from use of polluting fuels for cooking negatively affects the health of millions of individuals, particularly in low- and middle-income countries. As fine particulate matter (PM₂.₅) exposures from HAP vary over the course of a day, we examined cooking environment and socioeconomic factors associated with real-time fluctuations to PM₂.₅ improve exposure assessment in future epidemiological analyses.[¤]METHOD[|]We used machine learning (ML) models to predict real-time (1) PM₂.₅ kitchen concentrations and (2) PM₂.₅ cook exposures obtained from the CLEAN-Air(Africa) programme, which was conducted across three peri-urban communities in sub-Saharan Africa: Mbalmayo, Cameroon; Obuasi, Ghana; and Eldoret, Kenya. Household survey variables (e.g. number of times cook left home; walking time to nearest major road) and real-time measurements (CO, temperature, relative humidity), were incorporated. ML models, including decision trees, random forest, XGBoost, support vector regression, K-nearest neighbors, and neural networks, were evaluated using the coefficient of determination (R²) and root mean square error (RMSE).[¤]RESULTS[|]Across all communities, spikes in geometric mean PM₂.₅ kitchen concentrations were observed during evening cooking hours (5-8 pm), with the highest spike in Eldoret (96-188 µg/m³). The XGBoost (R²=0.86; RMSE=229) and random forest models (R²=0.85; RMSE=243) performed best for prediction of PM₂.₅ kitchen concentrations. Spikes in geometric mean PM₂.₅ cook exposures were also observed during evening cooking hours (85-53 µg/m³). Model performance for cook exposures was similar to that of kitchen concentrations (XGBoost model (R²=0.84; RMSE=140); random forest (R²=0.83; RMSE=140)). Real-time CO concentrations, temperature, relative humidity, household size, primary stove temperature, and community of residence were the strongest predictors of real-time PM₂.₅ cook exposures and kitchen concentrations.[¤]CONCLUSIONS[|]‘Tree-based’ ML methods were the most robust for predicting real-time PM₂.₅ kitchen concentrations and cook exposures in peri-urban sub-Saharan Africa. ML is a promising tool for scaling up PM₂.₅ exposure measurements in future epidemiological research.[¤] | en_US |
dc.language.iso | en | en_US |
dc.publisher | EHP Publishing | en_US |
dc.subject | Air pollution | en_US |
dc.title | Using machine learning to predict real-time PM2.5 concentrations from household air pollution in peri-urban sub-Saharan Africa | en_US |
dc.type | Article | en_US |
Appears in Collections: | School of Medicine |
Files in This Item:
There are no files associated with this item.
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.