Air Quality Predication Using Machine Learning Methods Based on Monitoring Data

Abstract

This research introduces a novel methodology for air quality prediction that addresses the limitations of traditional Air Quality Index (AQI) forecasting models by leveraging machine learning and enhanced secondary data modeling. The dataset utilized includes both forecast and actual measurements of primary pollutant concentrations and meteorological conditions, collected from monitoring stations in Jinan, China, from July 23, 2020, to July 13, 2021. A comprehensive correlation analysis identified ten key meteorological factors influencing pollutant concentrations, assessed through univariate and multivariate techniques. Performance evaluation of various machine learning algorithms revealed the Decision Tree and Random Forest models achieving high accuracies of 99%. Additionally, the K-Nearest Neighbors (KNN) classifier also demonstrated an accuracy of 99%, while Logistic Regression showed a training accuracy of 72%. These findings affirm the reliability and efficacy of machine learning techniques in enhancing air quality forecasting and underscore the importance of selecting appropriate algorithms for accurate predictions.