Cybersecurity for the Internet of Things: An AI-Based Adaptive Model and Landscape Analysis — A Systematic Literature
TL;DR
This study develops machine learning–based AQI forecasting models using real-time monitoring data from the Central Pollution Control Board, which includes 18 environmental parameters collected between February 2023 and October 2025, and shows that the tuned CatBoost model delivered the best predictive performance.
Abstract
The Taj Trapezium Zone (TTZ), which includes the industrial areas of Firozabad and Mathura, was established to shield the Taj Mahal from the harmful effects of declining air quality. Accurate Air Quality Index (AQI) forecasting is crucial for executing the Graded Response Action Plan (GRAP) of the Commission for Air Quality Management, as it necessitates early warnings of pollution episodes. This study develops machine learning–based AQI forecasting models using real-time monitoring data from the Central Pollution Control Board, which includes 18 environmental parameters collected between February 2023 and October 2025. Four ensemble algorithms—AdaBoost, XGBoost, LightGBM, and CatBoost—were employed with TimeSeriesSplit validation, Optuna-based hyperparameter tuning, and extensive preprocessing. The findings show that the tuned CatBoost model delivered the best predictive performance, with R² values of 0.995 in Firozabad and 0.951 in Mathura. SHAP analysis highlighted site-specific pollutant influences, revealing that both PM2.5 and PM10 significantly impacted AQI variations in Firozabad, while PM10 was the primary factor affecting AQI levels in Mathura. The CPCB breakpoint methodology confirmed that the predicted AQI categories align with observed classifications, ensuring reliable GRAP decision support. By enabling 72-hour advance forecasts, the proposed models offer a practical early-warning framework for proactive pollution control, targeted emission management, and enhanced air quality governance within the TTZ region.