Modelling Malaysia air quality data using Bayesian Structural Time Series models

Abstract

Air pollution poses a significant threat to human health and the environment, especially in developing nations facing rapid industrialization, urbanization, and increased vehicle emissions. As cities and factories continue to grow, the air quality problem worsens, making it crucial to enhance the monitoring, testing, and forecasting of air quality. In this context, this study focuses on building air quality models using Bayesian Structural Time Series (BSTS) models to predict air quality levels in Malaysia. The BSTS model integrates three main techniques: The structural model, which employs the Kalman filter approach to model trend and seasonality components; spike and slab regression for variable selection; and Bayesian model averaging to estimate the best-performing prediction model while accounting for uncertainty. The study utilized air quality time-series data spanning two years, from June 2017 to July 2019, obtained from the Malaysian Department of Environment (DOE). The primary objective of this study was to forecast air quality and assess the effectiveness of the Bayesian structural time series analysis on air quality time-series data. The results indicated that the BSTS technique is capable of modeling air quality time-series data with high accuracy, effectively capturing seasonal and trend components. The seasonal component showed a repetition of weekly concentration patterns, while the local linear trend component showed a steady decline in PM10 and PM2.5 concentration levels in most stations. Regression analysis demonstrated that humidity and ambient temperature significantly affected air quality in most locations in Malaysia.