<![CDATA[Increasing air pollution levels worldwide have raised serious concern regarding environmental sustainability and public health, making systematic analysis of pollution patterns increasingly important. This study analyzes global air quality patterns using three unsupervised clustering algorithms, K-Means, DBSCAN, and Hierarchical Clustering, applied to a dataset of 23,463 records from 175 countries, utilizing four major pollutant indicators: CO, Ozone, NO₂, and PM2.5 AQI values. Prior to clustering, outlier handling was evaluated by comparing three approaches: original data, IQR, and Z-Score methods, with the original data selected based on the highest silhouette score of 0.5696. Cluster configuration analysis indicated that four clusters provided the most interpretable segmentation structure. K-Means clustering produced four interpretable air quality groups, namely Clean, Moderate, Unhealthy, and Hazardous, with PM2.5 identified as the most dominant pollutant based on its strong correlation with the overall AQI Value (r = 0.985). Among the three algorithms, Hierarchical Clustering achieved the highest Silhouette Score (0.6341), followed by DBSCAN (0.6161) and K-Means (0.5696). Despite this, K-Means was selected as the primary algorithm due to its scalability, computational efficiency, and interpretability. A stability test conducted on the K-Means model across 20 trials confirmed consistent clustering performance, with a mean Silhouette Score of 0.5491 and a standard deviation of 0.0193. External validation against EPA AQI categories yielded an ARI of 0.2010 and NMI of 0.2410, confirming partial alignment with internationally recognized standards. The findings demonstrate that unsupervised clustering provides an effective framework for identifying global air quality patterns and supporting evidence-based environmental policy decisions.]]>
Copyrights © 2026
