AIR QUALITY PREDICTION FOR INDIAN CITIES USING SPARTIO-TEMPORAL DEEP LEARNING MODELS
TL;DR
The results indicate that spatial dependency modeling significantly enhances predictive performance in urban air quality systems and reduces RMSE by 18–25% for 1-hour forecasting and 15–20% for 24-hour forecasting.
Abstract
Air pollution has become one of the most critical environmental and public health challenges in India. Major metropolitan cities such as Delhi, Mumbai, and Hyderabad frequently record hazardous Air Quality Index (AQI) levels due to vehicular emissions, industrial activities, meteorological variability, and urbanization. Accurate short-term and multi-horizon AQI forecasting is essential for early warning systems and policy intervention. This study proposes a comprehensive spatio-temporal deep learning framework integrating Long Short-Term Memory (LSTM) networks and Graph Neural Networks (GNN) for multi-city AQI prediction using Central Pollution Control Board (CPCB) data from 2018–2024. The model captures both temporal pollutant dynamics and spatial inter-city correlations. Comparative evaluation against ARIMA and Random Forest models demonstrates that the proposed hybrid model reduces RMSE by 18–25% for 1-hour forecasting and 15–20% for 24-hour forecasting. Statistical significance testing confirms robustness (p < 0.05). The results indicate that spatial dependency modeling significantly enhances predictive performance in urban air quality systems.