Air Pollutant Concentration Analysis and Prediction using AI

Air pollution, stemming from human activities like energy production and transportation, poses a significant environmental threat. It is important to predict the accurate concentration of various pollutants for proactive safety of public health. This study utilizes a dataset from a heavily polluted area in Italy, comprising hourly averaged responses from metal oxide sensors. Various models, including deep learning (LSTM and GRU) and machine learning (Linear Regression, Decision Tree, Random Forest, SVM, Huber Regressor, Gradient Boosting Regression, K-Neighbors Regression, AdaBoost Regression, Gaussian Regression, and MLP Regression), were employed to analyze pollutant levels, focusing on carbon monoxide (CO). Noteworthy patterns, such as elevated CO levels in October and distinct daily/weekly variations, were identified, offering a comprehensive understanding of air quality dynamics. The inverse correlation between ambient humidity and NO levels suggests links between pollutants and environmental variables. Among the tested prediction models, Gradient Boosting (GB) regression stood out, providing insightful and accurate results for pollutant concentrations and weekly variations. The refined GB model demonstrated exceptional reliability, with a notable test R2 score of 0.91, showcasing its effectiveness in capturing dataset intricacies.