Community-level education percentile rank estimation in China using multi-source big data and machine learning.

Area of Science:

• Socio-spatial analysis
• Geographic Information Systems (GIS)
• Machine Learning

Background:

• Limited availability of fine-grained socio-economic data in China hinders effective policymaking.
• Traditional measures like years of education are less accurate indicators of social status.
• There is a need for a high-resolution dataset on social status at the community level.

Purpose of the Study:

• To create an open-access, community-level dataset of education percentile rank for China.
• To provide a more accurate indicator of social status than traditional education metrics.
• To facilitate fine-grained socio-spatial analysis and inform policy decisions.

Main Methods:

• Utilized an XGBoost machine learning model to estimate education percentile rank.
• Integrated multi-source data including surveys, points of interest, road networks, and night-time lighting.
• Employed computer vision techniques (semantic segmentation, object detection, image regression) on street view imagery.

Main Results:

• Developed a dataset covering 122,126 communities, representing 97.9% of prefecture-level and 81.8% of county-level administrative units.
• Achieved high accuracy in education predictions at prefecture, county, and community levels.
• The dataset captures the relationship between social status and built environment characteristics.

Conclusions:

• The newly developed dataset provides a valuable resource for socio-spatial research in China.
• Enables detailed analysis of social disparities and their spatial distribution.
• Supports evidence-based policymaking by offering granular socio-economic insights.