Predictive Modeling for Forecasting Air Quality Index (AQI) Using Time Series Analysis

Abstract

Air pollution is a widespread problem in India. The study focuses on forecasting the air quality index (AQI) using time series modeling techniques for the most polluted area of Dehradun City in Uttarakhand state, India. The train test approach of machine learning and Akaike information criterion (AIC) have been used on the monthly data of five years to select the best auto-regressive model. Using the auto-correlation functions (ACF and PACF) and the seasonality component in the time-series dataset, a seasonal autoregressive moving average (ARMA) model with its minimum AIC has been chosen to forecast the AQI. This model is also validated by comparing its predicted values with the actual values of AQI. The results showed that the seasonal ARMA model of (1,0,0)(1,0,0)12 could forecast AQI based on a stationary dataset. The research also indicates that the asthma patients of the Himalayan Drugs-ISBT region may experience more health effects, especially in winter, due to poor air quality. The model can be helpful for a scientist and the government to take precautionary measures in advance.