Machine learning-enhanced assessment of urban sustainable development goals progress

Abstract

Assessing urban sustainable development performance is vital for advancing global sustainable development goals (SDGs), yet it's often hindered by insufficient statistical data. Here we established machine learning models with the consideration of autocorrelation feature to fill 27 % of the missing values, achieving an average R² of 0.83 in our developed urban SDG framework, which encompasses 117 indicators for 286 Chinese cities. Our findings reveal a notable enhancement in the overall sustainable performance of Chinese cities from 2001 to 2020, with heightened competition particularly evident among middle-ranked cities. However, the distribution of urban SDG Index scores unveils significant spatial heterogeneity; while inter-regional disparities are diminishing, intra-regional differences among cities are widening. Our results after post-upscaling show a strong correlation with previous comprehensive national studies that utilized more indicators. Additionally, they provide extra insights compared to prior urban-scale studies that employed a fewer indicators. These results can assist policymakers in discerning the performance of urban SDGs and formulating appropriate solutions.