Identification of urban-rural integration types in China – an unsupervised machine learning approach

Purpose - Development of urban-rural integration is essential to fulfill sustainable development goals worldwide, and comprehension about urban-rural integration types has been highlighted as increasingly relevant for an efficient policy design. This paper aims to utilize an unsupervised machine learning approach to identify urban-rural integration typologies based on multidimensional metrics regarding economic, population and social integration in China. Design/methodology/approach - The study introduces partitioning around medoids (PAM) for the identification of urban-rural integration typologies. PAM is a powerful tool for clustering multidimensional data. It identifies clusters by the representative objects called medoids and can be used with arbitrary distance, which help make clustering results more stable and less susceptible to outliers. Findings - The study identifies four clusters: high-level urban-rural integration, urban-rural integration in transition, low-level urban-rural integration and early urban-rural integration in backward stage, showing different characteristics. Based on the clustering results, the study finds continuous improvement in urban-rural integration development in China which is reflected by the changes in the predominate type. However, the development still presents significant regional disparities which is characterized by leading in the east regions and lagging in the western and central regions. Besides, achievement in urban-rural integration varies significantly across provinces. Practical implications - The machine learning techniques could identify urban-rural integration typologies in a multidimensional and objective way, and help formulate and implement targeted strategies and regionally adapted policies to boost urban-rural integration. Originality/value - This is the first paper to use an unsupervised machine learning approach with PAM for the identification of urban-rural integration typologies from a multidimensional perspective. The authors confirm the advantages of this machine learning techniques in identifying urban-rural integration types, compared to a single indicator.