AIrguardplus AI: An Intelligent AIr Pollution Forecasting System Using Ensemble Machine Learning

Air pollution has emerged as a defining environmental and public health challenge of the twenty-first century. This paper presents AirGuardPlus AI, an intelligent air quality prediction system that employs an ensemble of three independently trained machine learning models to forecast ambient Air Quality Index (AQI) values from real-time atmospheric sensor measurements. The system is built on the UCI Air Quality dataset comprising 9,471 hourly measurements recorded at a road-level monitoring station in an Italian city between March 2004 and February 2005. The ensemble architecture integrates three complementary approaches: a Random Forest regressor leveraging bagged decision trees for robust non-linear feature interaction modeling, a Gradient Boosting regressor that iteratively corrects prediction residuals through sequential ensemble construction, and a Long Short-Term Memory (LSTM) neural network that processes thirteen sensor inputs as a sequential signal to capture temporal dependencies. Predictions are produced by simple averaging across all three model outputs, reducing variance relative to any individual model. The complete system pipeline encompasses European-format data parsing, -200 sensor error code imputation using bidirectional filling, domain-informed feature engineering, 80/20 stratified data splitting, model serialization via joblib and HDF5, and a Streamlit-based web interface for real-time ensemble AQI prediction. Experimental evaluation demonstrates that the ensemble achieves an R score of 0.932 and a Mean Absolute Error of 0.76 g/m, outperforming all individual component models. The system provides a practical, open-source, and locally deployable demonstration of ensemble machine learning for proactive air quality management.