Real-Time AQI Estimation: A Smart and Lightweight AI-Based System Using Gas Sensors

Air quality Index (AQI) is significantly related to quality of life and is an important metric that carries information about the dangers of air pollution. Air pollution both indoors and outdoors may cause serious health conditions like respiratory disease, premature death, or even cancer. Different pollutants like PM<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2.5</inf>, PM<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">10</inf>, CO, NO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">x</inf>, NH<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</inf>, SO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf>, O<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</inf> are used to assess AQI. Earlier studies worked on forecasting or classifying the AQI, but the consistency of unseen data remains underexplored. This study presents a real-time AQI classification system leveraging gas sensors and lightweight models. We proposed a real-time AQI classification system following a two-step architecture: pollutant-specific data imputation and AQI estimation/classification. Several models— Linear regression(LR), Decision Tree(DT), Random Forest(RF), XGBoost, and Deep Neural Networks (DNN) were evaluated. RF outperformed LR with an R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> score of 0.98 in AQI value estimation. XGBoost also achieved the highest accuracy of 98.61% in classification, outperforming DT and RF, followed by 98.28% and 98.18% respectively. The model’s performance was increased by 1.5% after missing value imputation. Missing pollutant concentrations were predicted with LR, RF, and Feed Forward Neural Network (FNN) using meteorological variables (temperature, humidity, etc.), but not in the final prediction. This system utilizes different gas sensors (MQ-136, MQ-135, MQ-131, MQ-2, and PMS5003) integrated with an Arduino to collect, calibrate, and analyze air quality data. Our system provides a real-time, portable, and affordable solution for indoor and outdoor environments, especially in resource-limited settings.