K-Means Clustering and Prophet-Based AQI Pattern Mining and Short-Term Forecasting for Asian Countries (2022–2025)

Abstract

This paper uses regional trends in air pollution in Asia as a case to explore the limitation of existing methods of analyzing pollution, which fail to effectively temporal pollution patterns. This is achieved by using a dataset of 19,949 records of weekly pollution in 40 countries in Asia obtained from Kaggle covering from 2022 to July 2025. In this paper, the K-Means clustering approach and Prophet algorithm have been used to analyze air quality data from the countries in Asia. Five clusters of countries with similar normalized AQI values have been identified, and each cluster has had four months of pollution forecasting performed at the cluster level. The results show varying levels of effectiveness of this approach to forecasting pollution with R² scores between 0.454 and 0.921 and mean absolute error of between 4.48 and 48.92. Large and stable clusters have relatively higher accuracy than smaller clusters with the smallest clusters (Cluster 4 with one country -India) being highly volatile. From this, it can be concluded that despite temporally sparse data, some valuable insights like monsoonal dip in pollution in July and August can be derived.