Enhancing particulate matter prediction in Delhi: insights from statistical and machine learning models

Area of Science:

• Environmental Science
• Data Science
• Urban Planning

Background:

• Particulate matter (PM10 and PM2.5) poses significant health risks in urban environments.
• Delhi experiences dynamic air quality variations influenced by emissions and weather.
• Accurate modeling of PM levels is crucial for effective air quality management.

Purpose of the Study:

• To evaluate traditional time series and machine learning models for PM10 and PM2.5 prediction in Delhi.
• To establish a baseline of air quality variations and identify trends.
• To enhance predictive accuracy by incorporating exogenous variables.

Main Methods:

• Utilized seasonal decomposition for baseline air quality analysis.
• Employed time series models (ARIMAX, SARIMAX) and machine learning models (RF, SVM).
• Incorporated exogenous variables including other pollutants and meteorological data.

Main Results:

• PM10 levels showed an increasing trend at several stations, while PM2.5 levels decreased overall.
• The Support Vector Machine (SVM) model demonstrated superior accuracy in predicting PM levels.
• SVM achieved testing RMSEs of 12.48-67.22 µg/m³ for PM10 and 8.38-48.95 µg/m³ for PM2.5.

Conclusions:

• The SVM model, enhanced with exogenous factors, provides a robust approach to PM prediction in urban settings.
• Incorporating diverse exogenous variables significantly improves environmental modeling and forecasting capabilities.
• This research offers actionable insights for policymakers to advance urban air quality management.