Multi-Modal Deep Learning Framework for AQI Prediction: An Extensive Analysis of 15 Monitoring Stations in Hyderabad Using Historical and Real-Time Data

Especially in fast-expanding cities like Hyderabad, where industrial activity, traffic emissions, and weather conditions create difficult air pollution patterns, air quality forecasting is a serious issue in urban areas. Traditional approaches for predicting the Air Quality Index (AQI) sometimes fall short because they do not represent the non-linear relationships and time dependencies found in air quality information. The research proposes a multi-modal deep learning architecture using Long Short-Term Memory (LSTM), Transformer, and Random Forest models to study past air quality data from 15 monitoring stations in Hyderabad. With a Root Mean Square Error (RMSE) of 3.7, the framework beats conventional statistical techniques and one deep learning technique, with 94.65 percent accuracy. Among the most important contributions are the creation of a strong preprocessing pipeline, the identification of important features including PM2.5, temperature, and wind speed, and the validation of real-time processing capacity. Seasonal study shows major fluctuations, with winter months having PM2.5 levels 45% above the yearly average. In addition, the research emphasizes the practical effects of the framework for policy formulation and management of urban air quality by providing a scalable answer for real-time AQI forecasting and long-term trend analysis.

• Air quality prediction
• Deep learning
• Hyderabad
• Real-time data
• WAQI API