Design and synthesis of time series forecasting and deep learning prediction model for air quality index prediction in Indian cities

Abstract

Modeling air quality by considering the complexities of randomness in pollutant concentrations and meteorological parameters to forecast real-time Air Quality Index helps mitigate public health risks. The uncertainty of prediction exists due to the high-dimensional nature of predictor variables, making the design of an early warning system highly critical and challenging. With the aim of ensuring accurate AQI forecasts, a statistical time series forecasting model, namely Vector Auto Regression (VAR), and an artificial neural network-based Long Short-Term Memory (LSTM) are integrated to form the Fusion Forecasting Model (FFM), referred to as PrediCasting. The proposed PrediCasting FFM (PCFFM) has been tested with air pollutant and meteorological data collected from the Central Pollution Control Board website for three major Indian cities: Noida, Hyderabad, and Vishakhapatnam. This work provides a detailed analysis of forecasting Air Quality Index by considering the correlation between all factors of air pollutants and meteorological parameters. Results demonstrate that, on average, the proposed PCFFM model has reduced the Root Mean Square Error value by 13.18% and 29.07% for 7-days-ahead and 14-days-ahead forecasts, respectively. Compared to existing models, 7-days-ahead and 14-days-ahead forecasts reduce Mean Absolute Percentage Error (MaPE) on average by 0.187 and 0.222, respectively.