Defining Rural Types Nearby Large Cities from the Perspective of Urban–Rural Integration: A Case Study of Xi’an Metropolitan Area, China

Abstract

Urban–rural integration (URI) is essential to achieving sustainable development. However, the rural areas surrounding large cities typically have a large scale and significant differences in development conditions. It is necessary to formulate rural development policies by category to better promote the integrated development between urban and rural areas, stimulate rural vitality, and create more significant opportunities for rural development. This study constructs an evaluation system for rural areas under URI, using the Xi’an metropolitan area as a case study. A clustering algorithm enhanced by the random forest (RF)–principal component analysis (PCA)–partitioning around medoids (PAM) method is applied to evaluate rural integration comprehensively. Key findings in this study include the following: (i) URI should be decoupled from administrative divisions, considering the complex impacts of multi-town functional spillover; (ii) ecological environment, economic development, public service allocation, and construction land supply are key factors influencing URI; (iii) the overall URI index in the Xi’an metropolitan area presents a “high in the center, low in the east and west” pattern. The rural areas with high URI index are around Xi’an and Xianyang, while other cities show insufficient communication with neighboring villages; (iv) rural areas can be categorized into four types of integration: ecological, ecological–economic, ecological–social–spatial, and ecological–economic–social–spatial, which exhibit an outward expansion of layers and extension along the east–west axis in the spatial structure of integration. Finally, differential development policies and suggestions for promoting urban–rural integration are put forward because of the different types of rural villages. This paper provides a framework for formulating rural development policies, significantly deepening urban–rural integration.

1. Introduction

Urban–rural integration (URI) development aims to enhance the close connections between urban and rural areas, facilitating the healthy bidirectional flow of resources and providing direction for the collaborative development relationship between cities and villages [1,2]. From the 19th National Congress of the Communist Party of China in 2017, which proposed to “establish and improve the system and policy framework for urban-rural integration”, to the Third Plenary Session of the 20th Central Committee in 2023, which pointed out that “urban-rural integration is an inevitable requirement for Chinese-style modernization”, and further to the Central Committee of the Communist Party of China and the State Council issuing the “Comprehensive Rural Revitalization Plan (2024–2027)” in early 2025, emphasizing the need to “promote urban-rural integration development” and “advance comprehensive rural revitalization in a classified manner”, the development of URI in China has gained increasing attention. It has become an essential sustainable and collaborative development strategy in urban and rural areas.

URI results from the evolution of urban–rural relations through theory and practice. Its theoretical basis has experienced three main paradigm evolutions: first, Marxist dialectics emphasizes the existence and resolution of urban–rural contradictions [3]; second, the new economic geography model reveals the interactive relationship between urban and rural areas by quantifying factor flows [4]; and third, cross-border governance and transaction cost analysis under the perspective of institutional economics provide a new framework for understanding urban–rural integration [5]. Notably, performance theory in European urban–rural relations research offers a new perspective on deconstructing the quality of urban–rural interaction through a multi-dimensional performance evaluation system [6]. It breaks through the traditional urban–rural dualism and emphasizes viewing urban and rural areas as dynamic functional network systems. Urban–rural relations gradually shift towards resource allocation, factor flow, mutual influence between urban and rural areas, and integrated development.

Many previous studies have focused on the theme of URI, covering various aspects such as conceptual analysis and theoretical framework construction [7,8], current status evolution and impact analysis [9,10], and holistic evaluation [11]. In recent years, an increasing number of scholars have also researched the classification of rural types [12,13,14] to provide a basis for differentiated development path selection for different villages. For example, Europe has a rich rural classification experience [15], showing concern for the causes and effects of social changes and attaching importance to the heterogeneity between rural areas [16]. However, there is currently a lack of research analyzing rural type classification in the frontier areas of URI—peri-urban areas.

The areas at the interface of metropolitan and rural areas are the regions where the flow of urban and rural elements and spatial transformations are the most intense and crucial zones for URI development [17]. In the “Rural Revitalization Strategic Plan (2018–2022)”, villages are categorized into four types: aggregation enhancement, urban–rural integration, characteristic protection, and relocation and dismantling, without further subdivision, making it difficult to provide precise guidance for formulating development strategies tailored to peri-urban villages. Due to their conditions and varying degrees of influence from nearby urban developments, different villages often exhibit differences in economic, social, spatial, and ecological integration during the URI process. As Ji and Tian (2024) point out, villages’ spatial characteristics, land use patterns, and development potential vary significantly across regions [18]. Rural areas near big cities, as transitional zones, are transforming from passively accepting spillovers to actively restructuring functions. Different rural regions need differentiated development policies, but current policy-making lacks scientific classification criteria, making it hard to meet diverse rural development needs precisely. Therefore, formulating rural development plans for peri-urban areas and scientifically conducting rural classification from the perspective of URI is fundamental and essential.

The classification of rural types often requires a comprehensive and objective evaluation index system. Some scholars have already conducted rural classification studies based on the characteristics of specific research areas, selecting evaluation indicators that cover various dimensions such as natural resources [19,20], economic and industrial development [21,22], social demographics [11,23], and public services [24]. However, most of these indicators are static and lack an understanding of the dynamics and trends of village development. Additionally, most existing research on rural classification is based on administrative jurisdiction units [25,26,27], with little consideration of the impacts of urban–rural and rural–rural interactions across administrative boundaries on village development; this situation is more pronounced in peri-urban areas. This cross-administrative interaction is crucial for accurately assessing urban–rural integration and policy effectiveness, and ignoring it may weaken URI policy implementation. Therefore, it is evident that the construction of evaluation indicators for rural classification that consider both static and dynamic characteristics from the perspective of URI and empirical analysis is urgently needed.

To this end, this paper constructs an evaluation index system for rural classification under the perspective of URI, encompassing four dimensions: ecological, economic, social, and spatial, based on the existing literature. Using the Xi’an Metropolitan Region as the geographical foundation for this study, we selected 3804 peri-urban villages (i.e., villages exhibiting URI characteristics) as analysis samples. Subsequently, we calculated the villages’ dimensional and comprehensive URI indices, identified rural clusters based on the evaluation indicators, and clarified the clustering implications in conjunction with the dimensional characteristics. Finally, we propose differentiated development strategies for villages in this region to promote the realization of URI and rural revitalization.

2. Study Area, Data, and Measurements

2.1. Study Area

Rural areas within metropolitan regions or urban agglomerations are often simultaneously influenced by functional spillovers from multiple cities. As the core city of the Guanzhong Urban Agglomeration, Xi’an exerts its influence on rural areas beyond its municipal administrative boundaries, necessitating an analysis of urban–rural integration within a broader regional context. Therefore, this study established a research area based on the Xi’an Metropolitan Region, from which suitable villages were selected for analysis. This region encompasses parts of Tongchuan to the north, Weinan to the northeast, and Xianyang to the northwest, spanning four prefecture-level cities (Figure 1).

Within this defined scope, selecting villages potentially exhibiting urban–rural integration characteristics surrounding Xi’an was conducted based on established methodologies. The urban–rural gradient division method proposed by Li Ruipeng et al. [28] and the population density threshold approach developed by Zhou Xiaochi et al. [29] were employed as primary references. Villages with impervious surface coverage ranging from 8% to 60% were categorized as peri-urban villages (Figure 2a). These areas were subsequently overlaid with regions exhibiting population densities between 1000 and 5000 persons/km² (Figure 2b). The preliminary selection was further refined through spatial optimization, incorporating considerations of population distribution patterns (Figure 2c) and arable land configurations (Figure 2d), while maintaining the principle of spatial continuity. This methodological process identified 3804 villages within the Xi’an Metropolitan Region as the final study samples (Figure 2e).

2.2. Data Collection and Pre-Processing

The data employed in this study were primarily categorized into three domains: socio-economic indicators, land use and built environment, and topographic features. For socio-economic analysis, Point of Interest (POI) data from 2012 to 2022 were acquired from Baidu Maps (https://map.baidu.com/) and Amap (https://ditu.amap.com), with particular emphasis on the scale and growth patterns of enterprises, healthcare facilities, and public infrastructure. Night-time light data were obtained from Chen et al.’s continuous time-series dataset of Chinese regions from 1992 to 2023, derived through DMSP-OLS and SNPP-VIIRS algorithms [30], which effectively captured the temporal variations in night-time illumination. GDP data were sourced from Zhao et al.’s predictive model that integrated night-time light time series and population imagery to estimate China’s GDP [31]. Electricity consumption data were derived from Chen’s high-resolution 1 km × 1 km grid data generated through spatial downscaling, providing insights into energy utilization patterns [32]. Age-specific population data were extracted from the 2020 Constrained Individual Countries dataset, developed by Bondarenko M et al., which offered global population estimates at the grid square level based on the Built-Settlement Growth Model, with detailed demographic breakdowns by gender and age groups. Additionally, mobile base station data were collected from OpenCelliD.

The foundational data were derived from multiple authoritative sources regarding land use and the built environment. The primary dataset included China’s first 1 m resolution national-scale land cover map, which was developed by Li et al. [33] through the integration of open-access remote sensing data. Additionally, the urban built-up area dataset for Chinese cities in 2020 established by Zhongchang Sun et al. [34] and the building vector data generated by Shi et al. [35] through a comprehensive large-scale mapping framework were incorporated. Furthermore, the road network data were obtained from the Open Street Map (OSM).

For topographic characterization, slope gradient and surface relief data were obtained from the European Space Agency. Ecological-related foundational data included the Normalized Difference Vegetation Index (NDVI) sourced from NASA Earth Data, along with the high-resolution, high-quality PM2.5 dataset for China (2000–2023) developed by Wei Jing et al. [36].

All fundamental data were aggregated and statistically analyzed at the village-level analytical unit. The data pre-processing was conducted using the min–max normalization method in ArcGIS Pro 3.0.1.

2.3. Measurements

Element flow is the foundation of URI. The flow space theory suggests that “flow space” refers to the location where various flow elements (such as people, capital, goods, information, and technology) exist and move, which, when mapped onto geographical space, forms a flow network composed of nodes such as cities, regions, and even countries [37,38]. From this perspective, URI is reflected as the spatiotemporal flow of elements that transforms the heterogeneous dual structure of urban and rural areas into a homogeneous unified structure, ultimately harmonizing economic, social, spatial, and ecological dimensions.

Guided by this theoretical foundation, this paper constructs an indicator system covering four dimensions: (i) economic integration is represented by night-time light index and electricity consumption to reflect economic vitality, GDP and the number of enterprises to measure economic strength, and the growth rate of the number of enterprises to indicate industrial development potential; (ii) social integration is analyzed through population size, aging rate, and the proportion of the labor force to understand population structure, and public service facilities’ equalization is reflected by the number of primary and secondary schools per thousand people, medical facilities and their growth, and infrastructure construction is shown through mobile base stations and public facilities; (iii) spatial integration emphasizes the territorial continuity between urban and rural areas, measured by the amount and growth rate of construction land, building density to assess spatial integration, and road network density to reflect the spatiotemporal “compression” effect; and (iv) ecological integration is assessed through slope and surface roughness to evaluate ecological foundations, with vegetation coverage and PM index reflecting environmental quality. The selection of indicators balances data availability with a combination of static and dynamic principles, aligning with the core aspects of flow space theory. The main considerations for our indicator selection are as follows:

(i)

Economic integration. Due to its high data accuracy, the night-time light index, a key indicator for economic activity intensity and spatial differences in urban–rural development, can capture informal economic activities and infrastructure distribution. The night-time light data are crucial for evaluating economic integration in data-scarce rural areas [39]. GDP reflects regional economic size, and the number of enterprises indicates industrial spatial agglomeration. Both are core indicators for assessing urban–rural economic integration [40].

(ii)

Social integration. Population indicators and labor force size are directly linked to social service needs, a significant factor in urban–rural social integration [41]. Indicators related to educational and healthcare facilities measure public service equalization, a prerequisite for narrowing the urban–rural welfare gap [42].

(iii)

Spatial integration. We chose the proportion and growth rate of construction land. Land use changes reflect spatial integration intensity and a high construction land growth rate signifies urban and rural land expansion demands [43].

(iv)

Ecological integration. The vegetation coverage index, acting as a proxy for ecosystem service provision, assesses the environmental condition of urban and rural systems [44].

Furthermore, the dynamic index selection spans from 2012 to 2022 for two main reasons. The first is data availability; socio-economic and geographic data have had reliable official sources since 2012. The second is policy continuity. The urban–rural integration strategy proposed at the 18th National Congress of the Communist Party of China in 2012, along with the 2014 National New-Type Urbanization Plan and the 2017 Rural Revitalization Strategy, forms a cohesive policy framework. These policies have driven long-term, in-depth urban–rural integration. The details of specific indicators of each dimension are in

Table 1. Urban–rural integration measurement indicator system.

Dimensions	Factors	Indicators	Calculation Methods	Remarks	References
Economic	Economic vitality	Night-time light index	GIS Zonal Statistics as a table tool to obtain ALL values	Static	[45,46]
		Electricity consumption	GIS Zonal Statistics as a table tool to obtain ALL values	Static	[47,48]
	Economic strength	GDP	GIS Zonal Statistics as a table tool to obtain ALL values	Static	[49,50]
		Number of enterprises	Total number of enterprise POI points	Static	[51,52]
		Enterprise growth	POI points (2022)–POI points (2012)	Dynamic
Social	Social structure	Population size	GIS Zonal Statistics as a table tool to obtain ALL values	Static	[53,54]
		Aging rate	Population aged ≥60/Total population	Static	[55,56]
		Proportion of the labor force	Population aged 15–64/Total population	Static	[57,58]
	Social security	Distribution of healthcare facilities	Statistical total number of healthcare POI points after hierarchical accessibility analysis	Static	[59,60]
		Growth in healthcare facilities	POI points (2022)–POI points (2012)	Dynamic
		Number of primary and secondary schools per thousand people	Number of primary and secondary schools/School-age population (6–18 years)/1000	Static	[41,61]
	Social infrastructure	Number of mobile base stations	Total number of mobile base station POI points	Static	[62,63]
		Number of public facilities	Total number of public service facility POI points	Static	[24,64]
		Growth in public facilities	POI points (2022)–POI points (2012)	Dynamic
Spatial	Urban spatial expansion	Growth rate of construction land	Rural construction land (2022)–Rural construction land (2012)/Rural construction land (2012)	Dynamic	[41,65]
		Proportion of construction land area	Construction land area/Total land area	Static
	Intensity of spatial development	Road network density	Total length of road centerlines/Land area	Static	[66,67]
		Building density	Area of building outline/Total land area	Static	[68,69]
Ecological	Terrain flatness	Terrain slope	Area with slope > 15°/Total area	Static	[70,71]
		Surface roughness	GIS Zonal Statistics as a table tool to obtain mean values	Static
	Ecological environmental quality	Vegetation coverage	GIS Zonal Statistics as a table tool to obtain mean values	Static	[72,73]
		PM2.5	GIS Zonal Statistics as a table tool to obtain mean values	Static	[74,75]
		PM10	GIS Zonal Statistics as a table tool to obtain mean values	Static

.

3. Methodology

3.1. Study Framework

Figure 3 illustrates the research framework of this study, which primarily consists of four steps. First, we collected and pre-processed multi-source data as the basis for sample selection and indicator extraction. Next, we identified the villages included in this study through multi-layer overlay analysis. Then, based on relevant theories of URI (such as the theory of factor flow), we developed a URI development evaluation indicator system encompassing four dimensions: economy, society, space, and ecology. A combined weight method was employed to derive the comprehensive index of URI. Simultaneously, an enhanced algorithm combining random forest–principal component analysis–partitioning around medoids (RF-PCA-PAM) was used for village clustering calculations. Finally, based on the radar chart of the fractal URI index, we synthesized the clustering characteristics, identified rural types, and proposed differentiated development strategies and recommendations in conjunction with the spatial distribution characteristics of the clusters. All these analyses were conducted in R 4.2.3.

3.2. Study Method

3.2.1. Weighting Methods

(1)

Analytic Hierarchy Process (AHP)

AHP is a multi-criteria decision-making approach that involves creating a hierarchical model consisting of three levels: the goal, criterion, and solution [76]. The calculation process is outlined as follows:

• Establish a weighted evaluation model based on evaluation indicators of urban–rural integration.
• Construct a judgment matrix using the Saaty 1–9 scale method, represented as $A=\{A_1,A_2,\cdot \dots ,\cdot A_n\}$.
• Perform a consistency check on the judgment matrix.
• Compute the subjective weight $W_{1j}$ for the j-th evaluation indicator.

$$I_C={\displaystyle \frac{{\lambda }_{mx}-n}{n-1}}$$

(1)

$$I_{CR}={\displaystyle \frac{I_C}{I_R}}$$

(2)

where $n$ is the order of the matrix, ${\lambda }_{max}$ is the most significant or principal eigenvalue of the matrix. $I_C$ denotes the consistency index, an $I_{CR}$ is the randomized consistency test.

$$W_{1j}={\displaystyle \frac{1}{n}}{\displaystyle \sum _{j=1}^n\frac{{\alpha }_{jk}}{{\displaystyle \sum _{k=1}^n{\alpha }_{jk}}}}$$

(3)

where $j$ = 1, 2, 3, …, n, and $a_{jk}$ represents the relative scale of indicator j to indicator k.

(2)

Entropy Weight Method (EWM)

The essence of entropy is the degree of internal chaos in a system [77]. The entropy method quantifies the uncertainty and variability of indicators by measuring the amount of information through information entropy [78]. The method removes the impact of human factors on subjective weight assignment and helps prevent information overlap among multiple indicators. The calculation process is outlined as follows:

• Calculate of the share of village $i$ under indicator $j$ in the calculation of the indicator: in which $i=1,2,\cdots ,m,j=1,2,\cdots ,n$.

  $$Y_{ij}={\displaystyle \frac{X_{ij}}{{\displaystyle \sum _{i=1}^m{X}_{ij}}}}$$

  (4)
• Normalize the indicators.

  $$P_{ij}={\displaystyle \frac{x_{ij}^{\prime }}{{\displaystyle \sum _{i=1}^m{x}_{ij}^{\prime }}}}$$

  (5)
• Calculate the entropy $e_j$ of the j-th indicator based on the normalization matrix $Y=(y_{ij})m$: in which $k={\displaystyle \frac{1}{\mathit{ln}(m)}}$.

  $$e_j=-k{\displaystyle \sum _{i=1}^m{P}_{ij}}ln{P}_{ij}$$

  (6)
• Calculate the j-th indicator’s entropy weight.

  $$W_j={\displaystyle \frac{1-e_j}{{\displaystyle \sum _{j=1}^n(1-e_j)}}}$$

  (7)

(3)

Composite Weighting Based on Game Theory

In order to obtain an optimal solution, this study applied game theory principles [79], treating subjective and objective weights as the two opposing parties in the game. The combination coefficients were derived by minimizing the total deviation between the final weights and the subjective and objective weights, resulting in comprehensive weights that balance the strengths of both subjective and objective aspects. The calculation process is outlined as follows:

• Formulate a fundamental set of weight vectors.

By applying both subjective and objective weighting methods to $m$ indicators, we obtain the weight set $W^{\prime }=\{{\omega }_1,{\omega }_2\}$. Subsequently, any linear combination of these two vectors can be represented as $W$:

$$W=a_1{\omega }_1^T+a_2{\omega }_2^T$$

(8)

where $a_1$ and $a_2$ denote the combination coefficients for the subjective and objective weights, respectively.

ii

Optimize $a_1$ and $a_2$.

Optimize $a_1$ and $a_2$ to minimize the total deviation between the weight vector $W$ and ${\omega }_1$, ${\omega }_2$.

$$min({\displaystyle \sum _{i=1}^2\Vert {\displaystyle \sum _{j=1}^2{a}_j}{\omega }_j^T-{\omega }_i^T{\Vert }_2})$$

(9)

Based on the properties of matrix differentiation, in order to satisfy the above equation, its first-order derivative must satisfy the following linear equation:

$$\left[\begin{array}{cc}{\omega }_1{({\omega }_1)}^T& {\omega }_1{({\omega }_2)}^T\\ {\omega }_2{({\omega }_1)}^T& {\omega }_2{({\omega }_2)}^T\end{array}\right]\left[\begin{array}{c}{a}_1\\ a_2\end{array}\right]=\left[\begin{array}{c}{\omega }_1{({\omega }_1)}^T\\ {\omega }_2{({\omega }_2)}^T\end{array}\right]$$

(10)

iii

Obtain the ultimate optimal combined weight vector, $W$.

$$W={\displaystyle \sum _{i=1}^2\left({\alpha }_i{\omega }_i^T/{\displaystyle \sum _{j=1}^2{\alpha }_j}\right)}$$

(11)

(4)

Comprehensive index of URI

After obtaining the final weights from the game theory, normalized data were used to calculate the corresponding indices for URI development evaluation. These included the economic integration index $E$, social integration index $S_o$, spatial integration index $S_p$, and ecological integration index $E_c$ for each dimension. The calculation formulas are as follows:

$$E_i={\displaystyle \sum _{j=1}^n{W}_j}{X}_{ij}$$

(12)

where $E_i$ is the economic integration index of the $i$-th village; $W_j$ and $X_{ij}$ represent the weight and normalized value of the $j$-th indicator and $i$-th village, and $n$ is the number of villages. The calculation formulas for the social integration index $S_o$, spatial integration index $S_p$, and ecological integration index $E_c$ are the same. After obtaining the dimension indices, the comprehensive index $U$ is calculated based on the corresponding weights:

$$U_i=E_i{W}_E+S{o}_i{W}_{So}+S{p}_i{W}_{Sp}+E{c}_i{W}_{Ec}$$

(13)

3.2.2. Enhanced Clustering Method

We employed an unsupervised random forest (RF) algorithm to obtain the proximity matrix between the computation samples, subsequently incorporating principal component analysis (PCA) to capture the intrinsic structure of the data. Finally, clustering analysis was conducted based on the dimensionality-reduced indices.

(1)

Random Forest and Adjacency Matrix

The random forest model, introduced by Breiman [80], is a machine learning algorithm that combines multiple classification trees to overcome the instability and overfitting issues typically associated with a single decision tree. The model is robust and has strong generalization capabilities, enabling it to handle many input features and data samples without easily overfitting [81]. It does not require assumptions about the data following a specific probability distribution or being generated from a particular model. Additionally, it demonstrates high interpretability and tolerance for data outliers and noise, effectively avoids multicollinearity issues, and can assess the importance of each feature, providing reliable predictive performance. The equation for the model is shown as follows:

$$i{m}_i={\displaystyle \frac{1}{nt}}{\displaystyle \sum _{v\in S_{xi}}G}ain\left(X_i,v\right)$$

(14)

The adjacency matrix is used in graph theory to represent a graph. The adjacency matrix represents the edges between nodes. The adjacency matrix is a square matrix whose elements indicate the connectivity between nodes in the graph. The adjacency matrix is as follows:

$$A=\left[\begin{array}{ccc}0& 1& 0\\ 1& 0& 1\\ 0& 1& 0\end{array}\right]$$

(15)

(2)

Principal Component Analysis (PCA)

Principal component analysis (PCA) was initially introduced by Porter [82] and later independently developed by Hotelling [83]. A linear transformation technique creates a new dataset from the original one. The central concept behind PCA is to reduce the dimensionality of a dataset while preserving as much of the variability in the data as possible. The mathematical foundation of PCA is primarily based on the following steps:

• Data Centering

  $$\overline{X}={\displaystyle \frac{1}{n}}{\displaystyle \sum _{i=1}^n{x}_{ij}},\hspace{1em}j=1,2,\dots ,d$$

  (16)

  $$X_{centered}=X-\overline{X}$$

  (17)

  where $\overline{X}$ is a mean vector, and where each column represents the mean of the corresponding feature.
• Calculate the covariance matrix.

  $$C={\displaystyle \frac{1}{n-1}}{X}_{centered}^T{X}_{centered}$$

  (18)

  where $C\in R^{d\times d}$ is a symmetric matrix, and $C_{ij}$ represents the covariance between feature $i$ and feature $j$.
• Calculate the eigenvalues and eigenvectors of the covariance matrix.

  $$Cv=\lambda v$$

  (19)

  where $v$ is the eigenvector of the covariance matrix, and $\lambda$ is the corresponding eigenvalue.
• Select the eigenvectors corresponding to the largest eigenvalues as the principal components.

(3)

Partitioning Around Medoid (PAM)

Partitioning around medoid clustering is a commonly used clustering algorithm in data analysis [84], developed by Kaufman and Rousseuw in 1987. The algorithm is based on the classical partitioning process of clustering. It initially selects k-medoids and then iteratively swaps the medoid with non-medoid objects, thereby improving the overall quality of the clusters [85]. The PAM algorithm is generally more robust than K-Means, particularly in the presence of noise or outliers. The PAM algorithm aims to partition the dataset into a pre-specified number of clusters, selecting each cluster’s most centrally located object as the cluster center. PAM clustering is primarily based on the following steps by Tagaram Soni Madhulatha [86]:

• Select the initial cluster centers: randomly select some objects from the dataset as the initial representative objects for the clusters.
• Assign data points to the nearest medoid: assign each remaining object to the cluster represented by the nearest centroid.
• Update the cluster centers: check if other points can serve as the new medoid for each cluster.
• Repeat steps until there is no change in the medoid.

4. Study Results

4.1. Comparison of Models’ Performance

Table 2. Comparison of PAM, PCA-PAM, and RF-PCA-PAM Models.

Models	Silhouette Coefficient	Calinski–Harabasz Index	Davies–Bouldin Index
Model 1: PAM	0.023	247.871	2.445
Model 2: PCA-PAM	−0.041	18,999.91	0.577
Model 3: RF-PCA-PAM	0.305	1820.026	1.261

Note: The Silhouette Coefficient measures the compactness of data points within clusters and the separation between clusters, with higher values indicating better performance [88]; the Calinski–Harabasz Index is based on the ratio of between-cluster to within-cluster variance, with larger values indicating better clustering performance [89]; the Davies–Bouldin Index measures the separation and compactness of clusters, with smaller values indicating better clustering performance [90].

describes the models’ overall performance before and after RF and PCA’s introduction. It can be observed that after incorporating the proximity matrix generated by the unsupervised RF and adding PCA, Model 3 achieved a significant increase in the Silhouette Coefficient, reaching 0.305, indicating a relatively good clustering effect (a Silhouette Coefficient between 0.3 and 0.5 suggests acceptable clustering performance) [87]. Model 2 exhibited a negative Silhouette Coefficient, indicating poor clustering performance, with samples likely being assigned to incorrect clusters. Compared to Model 1, Model 3 has a larger Calinski–Harabasz Index, suggesting better separation between clusters, while the Davies–Bouldin Index is more minor, indicating a higher degree of compactness within clusters. Therefore, the overall results of Model 3 are more reliable.

4.2. Descriptive Statistics by RF-PAC-PAM Approach

From

Table 3. Descriptive statistics of all indicators in this study.

First Level		Second Level		Third Level
					Descriptive Parameters				Weight Calculation
Dimensions	AHP-EWM	Factors	AHP-EWM	Indicators	Min	Max	Mean	SD	AHP	EWM	AHP-EWM
Social	0.3611	Social security	0.1561	Distribution of healthcare facilities	4.0000	43,990.0000	254.8159	993.2089	0.0870	0.0412	0.0782
				Growth in healthcare facilities	−5.0000	140.0000	1.6672	5.9902	0.0204	0.0481	0.0257
				Number of primary and secondary schools per thousand people	0.0000	76.9231	0.7676	3.0161	0.0558	0.0367	0.0522
		Social infrastructure	0.1348	Number of public facilities	0.0000	39.0000	0.5696	1.7943	0.0664	0.0381	0.0610
				Growth in public facilities	−2.0000	35.0000	0.4441	1.5407	0.0158	0.0485	0.0220
				Number of mobile base stations	0.0000	243.0000	1.3543	7.3710	0.0560	0.0342	0.0518
		Social structure	0.0702	Population size	6.0000	58,766.0000	2790.3829	4229.1381	0.0139	0.0457	0.0200
				Aging rate	0.1169	0.1923	0.1756	0.0134	0.0082	0.0474	0.0158
				Proportion of the labor force	0.6446	0.7846	0.7001	0.0221	0.0311	0.0486	0.0344
Economic	0.2868	Economic strength	0.2035	GDP	0.0011	4690.2358	102.1404	262.7823	0.0955	0.0425	0.0854
				Enterprise growth	−12.0000	642.0000	5.0814	20.0701	0.0676	0.0479	0.0638
				Number of enterprises	0.0000	642.0000	5.7538	20.9113	0.0576	0.0405	0.0543
		Economic vitality	0.0833	Electricity consumption	39,409.2285	22,134,927.0000	3,748,178.3831	5,544,468.6597	0.0546	0.0438	0.0525
				Night-time light index	0.2609	63.0000	23.7732	17.5534	0.0268	0.0476	0.0308
Ecological	0.2072	Ecological environmental quality	0.1512	Vegetation coverage	0.1512	0.6506	0.4689	0.0573	0.0893	0.0490	0.0816
				PM2.5	30.0750	53.9000	46.9016	3.2843	0.0431	0.0485	0.0441
				PM10	66.4667	106.2667	93.7368	5.5566	0.0200	0.0486	0.0255
		Terrain flatness	0.056	Surface roughness	0.9193	52.6171	3.9151	4.4094	0.0237	0.0491	0.0286
				Terrain slope	0.0000	0.0152	0.0014	0.0024	0.0223	0.0490	0.0274
Spatial	0.1449	Urban spatial expansion	0.0952	Proportion of construction land area	0.0001	0.9846	0.2140	0.1707	0.0564	0.0477	0.0547
				Growth rate of construction land	−0.9127	5.0400	0.1013	0.3100	0.0385	0.0489	0.0405
		Intensity of spatial development	0.0497	Road network density	0.0013	25.6163	3.1164	2.7693	0.0212	0.0472	0.0262
				Building density	0.0000	429.8823	0.1180	6.9754	0.0288	0.0012	0.0235

, it can be observed that the top 10 indicators, in order, are GDP, vegetation coverage rate, distribution of healthcare facilities, number of enterprises, number of public facilities, proportion of construction land area, number of enterprises, electricity consumption, number of primary and secondary schools per thousand people, and the number of mobile phone base stations, which reflect their significant impact on the degree of URI. According to the standard deviation data for each indicator, electricity consumption, population size, distribution of healthcare facilities, and GDP have more significant standard deviations, indicating a higher degree of dispersion, with significant differences between villages. Overall, the weights for the social, economic, ecological, and spatial integration indices decrease sequentially, at 0.3611, 0.2868, 0.2072, and 0.1449, respectively, indicating that their roles in the comprehensive URI degree decrease progressively.

4.3. Comprehensive and Dimensional URI Indices

Based on the combined weights analyzed in Section 4.2, further calculations and analysis of the URI degree of the sample villages were conducted. Figure 4 shows the spatial distribution of URI at each dimension and comprehensive level.

From a general perspective (Figure 4e), the spatial distribution of the comprehensive URI index shows a “high in the center, low in the east and west” pattern. Villages with the highest comprehensive indices are primarily located in the main urban area of Xi’an, the Gaoling District, and the built-up areas of Xianyang, as well as the central and northern parts of the Xi’an region. This reflects the close connections between these villages and Xi’an and Xianyang. In contrast, the integration levels of villages in the central areas of Weinan, Yangling, Fuping, Xingping, and Yanliang with surrounding villages are generally lower.

At the dimensional level, spatial integration (Figure 4c) and social integration (Figure 4b) generally show a pattern where the degree of integration increases the closer the area is to the core built-up areas of Xi’an and Xianyang. The spatial pattern of economic integration (Figure 4a) is similar to that of the comprehensive integration index, with the central region relatively higher than other areas, particularly Xi’an’s northern and central regions. Villages with higher ecological integration indices are mainly distributed in the west and northeast edges of the study area.

4.4. Clustering Distribution and Definition

4.4.1. Spatial Distribution of Clusters

We applied the RF-PCA-PAM method to obtain four clusters, with the orange (cluster 1), green (cluster 2), blue (cluster 3), and red (cluster 4) clusters containing 2093, 846, 603, and 262 villages. The spatial distribution of the clustering results is shown in Figure 5. The results reveal a typified layer structure that expands outward from the core urban areas of Xi’an and Xianyang. Additionally, urban and rural areas exhibit a continuous integration pattern in the east–west direction, demonstrating a particular integration axis. Specifically:

• In terms of the layer structure, the villages located in the central area belong to the comprehensive integration type (red cluster), mainly distributed between the main urban area of Xi’an City, the Gaoling District, and the built-up areas of Xianyang City, influenced by the radiative impact of urban functions. Surrounding the red cluster are villages with relatively good ecological–social–spatial integration (blue cluster); these villages are located near the main urban areas of Xi’an City, Xianyang City, and the Gaoling District, with favorable location conditions and frequent cultural, population, and material exchanges with cities. The third layer consists of villages with better ecological–economic integration (orange cluster), located on the outskirts of the main urban area, near county-level urban areas, and concentrated in regions surrounding the main urban area of Xi’an, such as the areas between Xingpin City, the Yanliang District, and Weinan City. The outermost layer comprises villages relatively far from the urban built-up areas, concentrated in the east and west, belonging to the green cluster.
• Along the east–west axis, the cluster of villages in the eastern yellow region has a larger contiguous area and a higher level of integration, closely linking Weinan City, Fuping County, the Yanliang District, and the central urban areas of Xi’an and Xianyang Cities. In contrast, the villages around the western areas of the Yangling District and Xingping City have lower degrees of contiguous clustering.
• Except for the main urban areas of Xi’an, Gaoling, and the built-up areas of Xianyang City, the degree of social and spatial integration between other urban core areas and surrounding villages is relatively low.
• Villages of the red and blue types show a few cases that are not adjacent to the main urban areas of Xi’an, the Gaoling District, and Xianyang City, indicating that these villages, while somewhat distant from the core urban areas, still maintain strong economic, social, or spatial connections.

4.4.2. Characteristics and Definition of Clusters

We further plotted radar charts based on the integration indices of each cluster in different dimensions, using the mean values to represent the overall characteristics and to define the clusters. From Figure 6, we can observe the following: (i) all clusters performed well in ecological integration, indicating that most villages have likely established good collaborative relationships with cities in terms of ecological protection and utilization; (ii) compared to clusters 2 and 3, cluster 1 shows better performance in the economic dimension, with an average integration index 0.0117 and 0.0008 higher than those of clusters 2 and 3; (iii) cluster 3 has higher social and spatial integration indices than both clusters 1 and 2; and (iv) all integration indices for cluster 4 are at a high level, with the spatial integration and social integration indices ranking first, and the economic integration index ranking second.

Based on the characteristics observed in the above radar chart, we define the four clusters as follows: ecological–economic integration type (cluster 1, orange), ecological integration type (cluster 2, green), ecological–social–spatial integration type (cluster 3, blue), and comprehensive integration type (cluster 4, red).

5. Discussion and Conclusions

Our research considers the inherent characteristics of rural areas and incorporates representative indicators of interaction and communication between urban and rural areas, establishing a comprehensive integration index evaluation system of peri-urban villages. Based on this framework, we employed an improved machine learning-enhanced classification algorithm for village categorization and utilized radar charts of dimensional features to determine clustering implications. This approach reveals the spatial pattern characteristics of URI and guides the formulation of differentiated development strategies for villages. This study identified significant heterogeneity in the degree of URI and its spatial distribution among 3804 villages surrounding the Xi’an metropolitan area. It also determined key influencing factors that promote rural development under urban–rural integration.

The functional radiation brought by large cities often impacts neighboring villages in an overlapping and complex multi-dimensional manner, a phenomenon commonly observed in urban agglomeration areas or urban corridors. The Growth Pole Theory posits that cities (growth poles) exert polarization and diffusion effects on surrounding villages [91], indicating that the relationship between urban and rural areas results from a composite network of mutual influences. From a performance theory perspective, these interactions dynamically shape rural socio-economic outputs. When policies align with local conditions, urban spillovers can boost rural industrial efficiency and labor allocation [92]. Consequently, the varying impacts of different cities in proximity may transcend administrative boundaries. Considering this characteristic, this study focuses on the Xi’an Metropolitan Region and incorporates the “urban-rural gradient” method to identify the villages included in the analysis. Unlike previous studies that classified villages based solely on singular administrative divisions [13,27,67], our empirical analysis conducted in Xi’an aligns more closely with objective realities, making the analytical results potentially more reliable.

Ecological conservation, industrial development, and infrastructure enhancement are crucial in promoting rural–urban integration by facilitating the flow of capital and population during rural revitalization. Firstly, ecological foundations constitute the fundamental basis for rural sustainability [93], while economic industries drive continuous rural development [94]. As emphasized by China’s “Two Mountains Theory”, “lucid waters and lush mountains are invaluable assets”. Peri-urban rural areas can actively develop ecological industries, such as eco-technological agriculture, green farming, and leisure tourism, with dynamic monitoring through indicators including the number of rural green enterprises and their growth rates and rural green vegetation coverage. Secondly, improving rural physical spatial environments is essential for enhancing living quality and supporting industrial development. Providing more industrial land and relevant preferential policies facilitates the attraction of urban enterprises, thereby promoting township industrial development [95,96,97]. Simultaneously, more pleasant rural living environments and comprehensive service facilities provide fundamental guarantees for people returning to rural areas for employment and residence [98]. These improvements also have the potential to attract potential urban populations, injecting new vitality into rural spatial revitalization [99,100,101].

This study found that villages in the central area of the Xi’an Metropolitan Region exhibit significantly higher levels of URI compared to those in the eastern and western parts of the region. The spatial structural characteristics are closely related to the urban system pattern of the metropolitan region. Xi’an and Xianyang, two major cities, have much larger scales and capacities than other towns; they are located centrally and considerably drive the development of surrounding rural areas. For example, villages between the two cities have formed a “near-airport and cultural-tourism” composite economic corridor, contributing to an annual output value growth of over 15% in 21 surrounding villages [102]. Villages with higher levels of social and spatial integration also tend to be closer to the city center, reflecting the frequent flow and interaction of urban and rural populations and the higher demand for better living standards among villagers. The increasing demand of urban residents for proximity to nature has made suburban tourism a vital means of relaxation and recreation [103]. Tourism-centered industry chain development has become essential for transforming peri-urban villages [104,105]. In contrast, economic integration does not require proximity to large metropolitan areas as much as the social and spatial dimensions. This may be related to Xi’an’s highly developed transportation infrastructure [106], the large-scale land demand for major economic industries, such as high-end manufacturing [107], and the attributes of external markets [108].

We found that rural clusters exhibit a general spatial pattern characterized by a central core centered around Xi’an and Xianyang, with a gradient diffusion effect. Similar rural spatial structure patterns have been found in Paris [109] and around central cities in Sweden [6]. The rural areas between the main urban area of Xi’an, Xianyang, and the Gaoling District of Xi’an displayed a multi-dimensional integrated state of economic, social, spatial, and ecological fusion. Critical infrastructure, such as Xianyang International Airport, and strategic policies, such as the national-level development zone, Xixian New Area, provide opportunities for the deep integration of urban and rural development in this region. The interaction and integration of capital, people, materials, and technology between urban and rural areas have fostered the vigorous development of rural areas [110], making this region a supporting rural hinterland for Xi’an’s “northward expansion” development [111]. An ecological–social–spatial integration type primarily characterized other villages adjacent to the main urban built-up areas. These areas were often more significantly impacted by urban disturbances [112], with intense changes in population mobility, rural space, and social network relationships [113,114]. Further outwards, there are rural areas with an ecological–economic integration type. These areas exhibited distinct contiguous patterns in the east–west direction, with the contiguous area in the eastern region being larger than that in the western region. The economic integration in these areas is also quite prominent, which may be related to the more developed township economies, the differentiated functional division between towns, and the spillover effects of coordinated development [115,116]. Some scholars have also observed varying degrees of integration and differentiation in the rural areas surrounding Wuhan [2,67,112]. However, the spatial distribution pattern they found differs from the results of this study, as it does not exhibit a concentric structure [67]. An interesting finding is that a few integrated rural areas show a “satellite” distribution, where they are spatially located at a certain distance from the built-up areas of large cities. The digital development of rural areas [117] and the growth in the number and quality of specialized enterprises [118] may have diminished the absolute importance of geographical location. For example, Huangliang Village has become a well-known “internet celebrity village” and art village based on its unique cultural development. It actively fosters and develops new rural industries, establishing a close market and leisure cultural experience with Xi’an.

Furthermore, we provide a practical evaluation framework and methodology to support the formulation of differentiated rural development policies. By systematically reviewing and summarizing the relevant literature on URI, we identified integration evaluation indicators from the perspective of rural development, attempting to construct an indicator system more suitable for classifying villages around metropolitan areas, which provides the foundation for the assessment. Then, we applied the RF-PCA-PAM combination for clustering analysis, which improves dimensionality reduction accuracy [119] and reduces data noise [120]. Our study found that the combined model outperforms the PAM and PCA-PAM models, contributing to more reliable evaluation results.

Based on the findings, we propose rural development policy suggestions to promote URI: (i) emphasize the current heterogeneity of URI, adopt differentiated development strategies by classification and region, and increase focus on rural areas with insufficient integration; (ii) focus on ecological environment protection, economic industries, local public service infrastructure, and construction land supply as key factors in promoting URI; and (iii) leverage non-traditional factors, such as rural e-commerce, to reduce the dominance of geographical location and enhance URI.

However, this study still has some limitations. First, the evaluation framework and methodology developed in this study were applied to the Xi’an Metropolitan Region, but further empirical analysis is needed to verify and obtain more universal research conclusions. Second, our study used a restricted set of dynamic indicators due to data availability limitations. Future research could reasonably incorporate more dynamic indicators to better reflect the changes urban–rural integration brought to rural areas. Third, due to the limitation of the applicability of PAM in extensive sample analysis [121], the research method in this study needs to be further developed. Additionally, some machine learning algorithms, such as extreme gradient boosting (XGBoost), have advantages in avoiding overfitting and improving analytical precision [122] and may be more suitable for classification based on semi-supervised learning.

Overall, previous research has rarely focused on the differences in types between peri-urban and hinterland villages. Our classification of village types from the perspective of URI is an innovative research exploration. The rural evaluation indicators and methods based on URI established in this study and the empirical results from the Xi’an Metropolitan Region provide valuable experience for the differentiated development of villages in surrounding urban areas, offering a positive reference for achieving rural revitalization.