Cluster analysis of PM_2.5 pollution in China using the frequent itemset clustering approach

In recent years, severe air pollution has frequently occurred in China at the regional scale. The clustering method to define joint control regions is an effective approach to address severe regional air pollution. However, current cluster analysis research on the determination of joint control areas relies on the Pearson correlation coefficient as a similarity measure. Due to nonlinearity and outliers in air pollution data, the correlation coefficient cannot accurately reveal the similarity in air quality between different cities. To bridge this gap, we proposed a method to delineate spatial patterns of PM_2.5 pollution and regional boundaries of polluted areas using the frequent itemset clustering approach. The frequent itemsets between cities were first mined, and the support values were employed as interestingness metrics to describe the significance of similar variation patterns between cities. Then, the hierarchical clustering method was applied to identify appropriate areas for joint pollution control. The proposed clustering algorithm exhibits the advantages of not requiring model assumptions and a robustness to the outliers, which is a cost-effective approach to define joint control regions. By analysing urban PM_2.5 pollution in China from 2015 to 2018, we obtained results demonstrating that the frequent itemset clustering approach can efficiently determine pollution patterns and can effectively identify regional divisions. The clustering approach could facilitate a greater understanding of PM_2.5 spatiotemporal aggregation to design joint control measures among areas. The findings and methodology of this research have important implications for the formulation of clean air policies in China.