Air quality index classification based on machine learning techniques: A case study from Hangzhou City

Abstract

Atmospheric pollutants pose a significant threat to human health and ecological stability, necessitating the development of accurate and reliable classification models to support informed decision-making. This study leverages state-of-the-art machine learning techniques to classify air quality levels in Hangzhou, a rapidly urbanizing city, in 2023. Through comprehensive data analysis, it examines the impact of data preprocessing and feature engineering on classification performance. It evaluates the effectiveness of five machine learning models: Random Forest (RF), Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), CatBoost, and Long Short-Term Memory (LSTM) networks. To enhance data integrity, preprocessing techniques such as anomaly detection, missing value imputation, and the Synthetic Minority Oversampling Technique (SMOTE) were applied. At the same time, feature engineering strategies, including seasonal and temporal feature extraction, were implemented to improve classification accuracy. Experimental results show that Random Forest achieved the highest classification accuracy and PR-AUC, surpassing XGBoost and CatBoost, demonstrating strong performance. While LSTM benefited from seasonal feature integration, its classification effectiveness remained lower than that of ensemble models, particularly in handling class imbalances. SVM exhibited moderate performance but struggled with complex data distributions. These findings highlight the strengths and limitations of each model, offering a scalable classification framework that can support urban air quality management and be adapted to similar environments.