Regression and multivariate models for predicting particulate matter concentration level

Abstract

The devastating health effects of particulate matter (PM₁₀) exposure by susceptible populace has made it necessary to evaluate PM₁₀ pollution. Meteorological parameters and seasonal variation increases PM₁₀ concentration levels, especially in areas that have multiple anthropogenic activities. Hence, stepwise regression (SR), multiple linear regression (MLR) and principal component regression (PCR) analyses were used to analyse daily average PM₁₀ concentration levels. The analyses were carried out using daily average PM₁₀ concentration, temperature, humidity, wind speed and wind direction data from 2006 to 2010. The data was from an industrial air quality monitoring station in Malaysia. The SR analysis established that meteorological parameters had less influence on PM₁₀ concentration levels having coefficient of determination (R ²) result from 23 to 29% based on seasoned and unseasoned analysis. While, the result of the prediction analysis showed that PCR models had a better R ² result than MLR methods. The results for the analyses based on both seasoned and unseasoned data established that MLR models had R ² result from 0.50 to 0.60. While, PCR models had R ² result from 0.66 to 0.89. In addition, the validation analysis using 2016 data also recognised that the PCR model outperformed the MLR model, with the PCR model for the seasoned analysis having the best result. These analyses will aid in achieving sustainable air quality management strategies.