Generalized Space-Time Autoregressive (GSTAR) for forecasting air pollutant index in Selangor

Abstract

This study presents the Generalized Space-Time Autoregressive (GSTAR) model, a multivariate time series approach that integrates spatial and temporal observations for data forecasting. This study's primary objective is to develop and apply the GSTAR model to forecast the Air Pollutant Index (API), which exhibits spatial-temporal dependencies between locations and time. Three areas in Selangor have been used in this study: Banting, Petaling, and Shah Alam. The model employs uniform and inverse distance weights to consider spatial relationships. The forecasting performance is assessed using Root Mean Square Error (RMSE). Although both weight methods yield comparable results, the GSTAR model with inverse distance weight is promising for API data forecasting with consistently low RMSE values. The result of this study emphasises the significance of location-based information in generating more efficient and informed solutions.