Performance and applicability of low-cost PM sensors to assess global pollution variability through machine learning techniques
Performance et applicabilité des capteurs à bas coût pour les particules fines (PM) dans l’évaluation de la variabilité mondiale de la pollution à l’aide de techniques d’apprentissage automatique
Résumé
Air quality monitoring and analyses became easy and affordable due to emergence of low-cost sensors. Recently, the efforts to improve the monitoring and understanding of region-specific air pollution events attracted immense global attention. Nevertheless, the applicability issues were observed due to data reliability and inconsistency, caused by reserve testing of performance parameters for better accuracy, selection and deployment of sensors without considering their fitness for the purpose, and area-specific requirements. This paper analyses and evaluates low-cost sensor deployments across lower, middle, and higher income group of countries, emphasizing variations in pollutant sources, performance parameters, and machine learning approaches for local source categorization. The performance parameters were analyzed using three Key parameters: (1) the Performance Index, (2) Sector Sensitivity Ratio, and (3) Data Reliability Indicator, that provide a comprehensive understanding of sensor efficiency in diverse environments. Our findings reveal distinct trends among income group countries. Higher income group countries exhibited the highest performance Index (0.35), followed by middle (0.33) and lower income group countries (0.27). However, the lower income group countries showed the highest data reliability indicator for maximum sector contribution (14.26), surpassing the higher (11.74) and middle income group (10.71) countries. Sector wise, transport (higher income), industry (middle income), and power (low income) demonstrated the highest data reliability based on its indicator. Additionally, it was observed that advanced machine learning algorithms helped to improve performance parameters, particularly in middle and lower income group countries where pollution variability is higher. These findings underscored the disparities in sensor performance and data reliability across diverse income groups.
Mots clés
- Generalized Additive Model
- Neural Network
- Gaussian Process Regression
- Support Vector Regression
- Multiple Linear Regression
- Linear Regression
- Support Vector Machine
- Artificial Neural Network
- Gradient Boosting Decision Trees
- Random Forest
- Low-cost sensor
- Particulate Matter
- Machine Learning
- Generalized Additive Model ML
- Neural Network GAM
- Gaussian Process Regression NN
- Support Vector Regression GPR
- Multiple Linear Regression SVR
- Linear Regression MLR
- Support Vector Machine LR
- Artificial Neural Network SVM
- Gradient Boosting Decision Trees ANN
- Random Forest GBDT
- Low-cost sensor RF
- Particulate Matter LCS
Domaines
| Origine | Publication financée par une institution |
|---|---|
| Licence |