Comparative Analysis of Event-Aware Air Quality Prediction Models

This study benchmarks machine learning and deep learning architectures — Random Forest, LSTM, CNN, and a hybrid CNN-LSTM model — for AQI prediction with a focus on event-aware modeling. Real-world events such as festivals, heavy traffic periods, and industrial activity are encoded as contextual features, enabling models to capture transient pollution spikes that standard approaches miss. Results demonstrate the superior capacity of hybrid deep learning models in handling temporal dependencies and event-driven anomalies in air quality time-series data.