Evaluation of Convenience of 15-Minute Community Life Circle Facilities and Analysis of Non-Linear Influencing Variables from the Perspective of Aging: A Case Study of Shenyang

Abstract

Amid rapid global population aging, developing age-friendly urban spaces centered on the “15-minute community life circle” has become a priority in planning research. Taking Shenhe District of Shenyang City, a region undergoing deep aging, as a case study, this research constructs a facility weighting system reflecting the actual needs of the elderly. Integrating multi-source spatial data, the XGBoost model and SHAP framework were applied to analyze the non-linear effects of socio-economic, functional, and land-use factors on facility convenience. Results indicate that: (1) facility convenience exhibits a distinct “west-high, east-low” spatial pattern, characterized by high agglomeration in the western core and significant deficits in the eastern fringe; (2) convenience levels vary across categories, with medical and health facilities showing the highest accessibility, while cultural and leisure (CALFs), life service, and elderly care service facilities (ECSFs) remain the primary deficiencies; and (3) influencing variables demonstrate complex non-linear mechanisms, wherein functional density and distance from the city center are critical drivers with non-monotonic effects, while road network density displays threshold effects, inhibiting ECSFs and CALFs at high densities. These findings provide a refined, quantitative basis for optimizing facility layouts and formulating urban renewal strategies to build age-friendly communities.

1. Introduction

In the 21st century, population aging has become an important social problem facing the whole world [1,2]. The United Nations projects that the population aged 60 and above will exceed 2 billion by 2050 [3,4]. China, home to the world’s largest elderly population [5], is expected to enter a stage of severe aging around 2035 [3]. This trend places significant strain on the supply of public services [6], such as elderly care and medical support [7]. It also challenges urban spatial planning [8] and resource allocation [9]. Following the UN Sustainable Development Goals’ (SDGs’) call for safe, inclusive, and accessible public spaces [10], creating convenient, livable environments for the aging population has become a priority in urban planning [11,12].

Previous studies have typically assessed public service supply and demand at the city [13], street [14], or grid scale [15]. In contrast, the “life-circle” scale is increasingly adopted as the fundamental unit for resource allocation [16,17], as it more accurately reflects service accessibility for the elderly and embodies a human-centric philosophy [16,18,19]. This concept originates from Clarence Perry’s “neighborhood unit” [20] and Jane Jacobs’ advocacy for mixed-use, walkable neighborhoods [21]. More recently, Professor Carlos Moreno introduced the “15-minute city,” [22] positing that residents should meet six core needs within a 15 min walk or ride. This concept serves as a direct theoretical reference for China’s “15-minute community life circle” (15 min CLC) [23]. Recently, the Chinese government issued the Standard for Urban Residential Area Planning and Design (2018) and the Community Life Circle Planning Guide (2021), establishing a “5–10–15 min” framework [24] for community service development [25].

The spatial supply and demand matching analysis of public service facilities has become common in academic research and planning practice, but when explaining the mechanism of residents’ convenient access to facilities, existing studies mostly adopt linear assumptions [26,27]. This approach often fails to capture real-world complexities. Accessibility is influenced by walking conditions [28,29,30], functional mixing [31], density [32], and socio-economic factors [33]. These variables interact dynamically rather than having simple additive effects. Relying on linear models may mask these complex mechanisms. Therefore, exploring non-linear relationships is essential for optimizing community planning.

Compared to other demographic groups, the elderly face restricted mobility and rely heavily on community-based public services [34,35]. Consequently, 15 min CLC serves as their primary sphere of activity [36]. As population aging intensifies, the development of these life circles remains constrained by practical challenges, including insufficient elderly service facilities, uneven spatial distribution, and inadequate user convenience. Enhancing the accessibility and convenience of facilities within this circle [37] is vital for improving the elderly’s quality of life and promoting healthy aging. Research at this scale not only deepens our understanding of public service supply patterns [17] but also provides direct strategies for optimizing urban service systems [38]. Previous studies have yielded valuable insights into specific facility types, such as medical and elderly care services, largely facilitated by the availability of scale-based indicators like bed counts and medical staff numbers [39,40]. Nonetheless, research has predominantly focused on isolated facility categories, often overlooking the integration of actual walking mobility and daily life contexts of older adults. A systematic comparison across different types of facilities remains limited, particularly at the community life circle scale, where non-linear influences on the convenience of commercial, daily service, and cultural facilities are yet to be examined. This research gap not only obscures how well the diverse needs of older adults are met, but also hinders the understanding of how socio-economic and urban form factors interact to shape facility service effectiveness.

Combined with the above analysis, the key to the planning of 15 min CLC against the background of aging lies not only in the balanced allocation of the number of facilities and spatial distribution, but also in the in-depth exposition of the complex mechanism that affects convenience. Taking Shenhe District of Shenyang City, a region undergoing deep aging, as a case study, this research integrates multi-source data (including points of interest (POI), residential areas of interest (AOI), and walking routes) to construct a convenience evaluation system based on the walk score (WS). Furthermore, the XGBoost model and SHAP framework are applied to analyze the non-linear impacts of socio-economic, functional, and land-use factors. By addressing the limitations of linear assumptions, this study provides a refined, quantitative basis for optimizing facility layouts and building age-friendly communities.

2. Materials and Methods

2.1. Study Area and Data

By 2025, China has fully entered the stage of moderate aging [39], with a large elderly population and accelerating aging speed. The demographic aged 65 and above has reached approximately 220 million, constituting 15.6% of the total population [40]. Among all provinces, Liaoning records the highest proportion at 21.9% [41], indicating a transition into severe aging. Shenyang, the capital of Liaoning Province, is located in the northeast of China. As the central city of the old industrial base, Shenyang has experienced a social and economic transformation over the past four decades, and the problem of population aging is particularly prominent. To actively address aging, the Shenyang government has expanded the coverage of convenience life circles. This initiative aims to optimize community elderly care and ensure that residents can access essential services within a 15 min walk, thereby promoting the equalization of public services.

Shenhe District (Figure 1) spans 58.95 km² and hosts a permanent population of 783,000. Elderly residents (aged ≥ 65) make up 16.81% of the population, exceeding the municipal average by 1.34 percentage points. The district administers 11 sub-districts, including Beizhan and Zhujianlu. As the fundamental grassroots administrative unit in Chinese cities, a sub-district typically covers at least 2 km² and serves a population of 50,000 to 100,000. Shenhe District exhibits a dual spatial structure. The western area is the old city center, characterized by a dense road network and mature public services. The residential buildings here are mostly multi-story apartments constructed between the 1980s and 1990s. In contrast, the eastern area is a newly developed zone. The road network is sparser but wider, and the housing stock consists primarily of modern high-rise residential complexes built after 2000.

Based on relevant references, according to the needs of the elderly [42,43,44], this research covers shopping, medical treatment, and leisure and travel scenarios, and follows relevant industry standards to screen elderly service facilities, and finally divides them into six categories and twenty-three sub-categories of elderly service facilities (

Table 1. Statistics on the number of elderly service facilities in the buffer zone.

Facility Category	Living Circle Hierarchy	Facility Subcategory	Number of Facilities	Stratified Statistics	Total Number
Convenience commercial facility (CCF)	5–10 min	Convenience store	525	557	11,557
		Vegetable market	32
	15 min	Restaurant	10,914	11,000
		Shopping mall, supermarket	86
Life service facility (LSF)	5–10 min	Express delivery point	59	59	1051
	15 min	Bank outlet	542	992
		Telecommunication outlet	255
		Neighborhood committee	156
		Police station	39
Elderly care service facility (ECSF)	5–10 min	Elderly dining point	94	152	201
		Day care center	58
	15 min	Nursing home	49	49
Medical and health facility (MAHF)	5–10 min	Pharmacy	864	1378	1481
		Clinic	514
	15 min	General hospital, community hospital	103	103
Cultural and leisure facility (CALF)	5–10 min	Elderly activity station	222	238	318
		Outdoor fitness venue, pocket park	16
	15 min	Elderly activity center	12	80
		University for the Elderly	6
		Community park, integrated park	46
		Square	16
Transportation facility (TF)	5–10 min	Bus stop	658	658	690
	15 min	Metro	32	32

). Residents at the boundary of Shenhe District can access facilities in adjacent districts. According to the Standard for Urban Residential Area Planning and Design [45], the radius of a 15 min life circle is typically 800–1000 m. To fully capture the actual usage of surrounding facilities, a 1500 m buffer zone was delineated outside Shenhe District. This zone defines the spatial scope of the study.

All online data sources were accessed in June 2025. POI data of public service facilities in Shenhe District and its 1500 m buffer zone and building vector data were extracted from Amap (https://ditu.amap.com/). Residential AOI data and road networks were respectively acquired from Baidu Map (https://map.baidu.com/) and OpenStreetMap (https://www.openstreetmap.org/). To ensure the quality of road network data, we performed topological checks and removed duplicates. We also manually verified the main roads against satellite imagery to ensure accuracy. Population data were sourced from the grid-based dataset shared by Chen et al. on Figshare (https://doi.org/10.6084/m9.figshare.24916140.v1) based on the 2020 census data, which offers superior accuracy compared to WorldPop and LandScan for China. Urban land use data were obtained from the dataset developed by Chen et al. (https://zenodo.org/records/15180905). Walking distances between residential communities and facility points were calculated using the Baidu Maps Walking Route Planning Application Programming Interface (https://lbsyun.baidu.com/faq/api?title=webapi/guide/webservice-lwrouteplanapi/walk), which reflects the actual walking path rather than Euclidean distance. Accounting for the physiological decline of older adults, walking speed was adjusted to 0.84 m/s based on the existing literature [46,47,48], yielding a 15 min CLC radius of approximately 760 m. To delineate the boundary of the 15 min community life circle, we utilized Mapbox’s Isochrone API (https://docs.mapbox.com/api/navigation/isochrone/) to generate 15 min walking isochrones starting from the geometric centers of each residential area AOI obtained in the previous text.

2.2. Research Framework

This study constructs a systematic analytical framework aimed at assessing the convenience of 15 min CLC for the elderly and revealing its influencing factors. The framework integrates multi-source geospatial data, an improved walk score, and interpretable machine learning models.

The data primarily include POI data for facilities and AOI data for residential points, as well as indicators such as building vector data, road networks, and population grids. Firstly, a modified walk score method is applied to quantify the facility convenience of each 15 min life circle as the dependent variable, followed by a comparative analysis of convenience levels across different facility categories and overall accessibility. Subsequently, predictive variables for each life circle are calculated across five dimensions: socio-economic conditions, construction scale, functional mix, transportation accessibility, and land-use structure. The XGBoost model is employed to capture the complex non-linear relationships between these variables and convenience. Finally, the SHAP method is applied to interpret the model results. The research framework is shown in Figure 2.

2.3. Research Methods

2.3.1. Evaluating the Convenience of 15 min CLC Based on Walk Score

The walk score method evaluates service accessibility by assigning facility weights based on user preferences while integrating distance decay and walking environment factors. Consequently, it accurately reflects the facility access level for target users, serving as a proxy for life circle convenience. The calculation involves three key steps [49,50,51]: (1) classifying facilities and establishing a weighting scheme; (2) calculating the base WS incorporating distance decay; and (3) deriving the final score by applying attenuation indices for intersection density and block length. Here, a “block” is defined as the area enclosed by roads. Block length refers to the length of the road segments forming the block boundaries. The formula is expressed as:

$$WS=\sum _{i=1}^n\{w_i\times f(d)\times [1-D(c)]\}$$

(1)

where $WS$ represents the walk score at a specific location; $w_i$ denotes the weight of the facility; $n$ indicates the facility type; f(d) is the distance decay coefficient corresponding to travel distance; and D(c) is the attenuation index corresponding to intersection density and block length (

Table 2. Comparison table of intersection density and block length attenuation rate.

Intersection Density (Intersections per Square Mile)	Attenuation (Penalty %)	Block Length (in Meters)	Attenuation (Penalty %)
>200	0	<120	0
150–200	1	120–150	1
120–150	2	150–165	2
90–120	3	165–180	3
60–90	4	180–195	4
<60	5	>195	5

Source: walkscore.com/Walk Score Methodology (https://www.walkscore.com/methodology.shtml).

).

(1)

Classifying facilities and establishing a weighting scheme

The survey was conducted among residents aged 60 and above in Shenhe District. To ensure that the results reflect regional characteristics, the survey covered all 11 sub-districts within the study area. Respondents were randomly selected in public spaces such as community centers and parks in each sub-district. This process yielded 467 valid responses from 483 returns (an effective response rate of 93.4%). The survey assessed facility usage frequency, usage diversity, and perceived importance. “Usage diversity”, representing the breadth of demand, is defined as the minimum number of facilities required to satisfy basic needs from near to far. Assuming equal access opportunities for facilities within the same category, respondents indicated the quantity (ranging from 1 to 10) needed to fully meet their demands. To better align with the actual needs of the elderly in China, this study adopts a localized optimization of the facility weighting method [50,52]. Unlike the standard approach, we determine the effective service quantity required for the majority. Therefore, the threshold was set to satisfy the needs of over 90% of respondents (

Table 3. Value of diversity in the use of elderly service facilities.

Facility Type	Certain Facilities Numbered from Nearest to Farthest										Value of Diversity (90%)
	1	2	3	4	5	6	7	8	9	10
Convenience store	53.42	30.43	9.32	3.10	1.86	1.24	0.62				3
Vegetable market	63.98	30.43	2.48	1.86	0.62	0.62					2
Restaurant	39.13	15.53	14.91	11.8	11.18	4.35	2.48	0.62			5
Shopping mall, supermarket	75.78	16.15	5.59	1.24	0.62	0.62					2
Express delivery point	73.29	24.22	1.86	0.62							2
Bank outlet	93.79	3.73	1.24	1.24							1
Telecommunication outlet	98.14	1.24	0.62								1
Neighborhood committee	95.65	3.10	1.24								1
Police station	97.52	1.86	0.62								1
Elderly dining point	55.90	38.51	2.48	1.86	1.24						2
Day care center	49.07	45.34	4.97	0.62							2
Nursing home	91.30	6.83	1.24	0.62							1
Pharmacy	70.81	19.88	4.35	3.10	1.24	0.62					2
Clinic	50.31	46.58	1.86	1.24							2
General hospital, community hospital	91.93	6.83	0.62	0.62							1
Elderly activity station	63.35	32.30	3.73	0.62							2
Elderly activity center	95.03	3.73	1.24								1
University for the elderly	92.55	6.83	0.62								1
Outdoor fitness venue, pocket park	55.28	40.37	3.73	0.62							2
Community park, integrated park	57.14	34.16	6.83	1.24	0.62						2
Square	94.41	4.97	0.62								1
Metro	68.32	27.33	3.10	0.62	0.62						2
Bus stop	52.80	38.51	6.21	1.24	0.62	0.62					2

). This approach avoids resource redundancy caused by extreme individual preferences and ensures a more efficient facility layout. For instance, over 90% of the elderly considered two elderly dining points sufficient.

While usage frequency reflects current behavioral characteristics, facility importance indicates latent demand. Synthesizing these dimensions provides a realistic basis for weight assignment. Following the determination of diversity values, the Expert Scoring Method was employed to establish the relative weights of “importance” and “frequency”. Using sum normalization, these weights were derived as 0.2 and 0.8, respectively. Subsequently, a weighted linear combination model was constructed to finalize the facility weighting table (

Table 4. Classification weight of elderly service facilities.

Facility Type	Classification Weight	Diversity Number	Diversity Weight
Convenience store	10.85	3	6.22	3.54	1.09
Vegetable market	14.05	2	9.52	4.53
Restaurant	3.52	5	1.49	0.59	0.57	0.45	0.43
Shopping mall, supermarket	5.22	2	4.30	0.92
Express delivery point	9.36	2	7.04	2.32
Bank outlet	1.65	1	1.65
Telecommunication outlet	0.92	1	0.92
Neighborhood committee	1.83	1	1.83
Police station	0.63	1	0.63
Elderly dining point	1.80	2	1.07	0.73
Day care center	0.96	2	0.50	0.46
Nursing home	0.88	1	0.88
Pharmacy	3.06	2	2.39	0.67
Clinic	0.97	2	0.50	0.47
General hospital, community hospital	1.87	1	1.87
Elderly activity station	2.72	2	1.80	0.92
Elderly activity center	1.11	1	1.11
University for the elderly	2.65	1	2.65
Outdoor fitness venue, pocket park	7.07	2	4.09	2.98
Community park, integrated park	10.95	2	6.85	4.10
Square	1.48	1	1.48
Metro	7.25	2	5.18	2.07
Bus stop	9.19	2	5.31	3.87

). The results demonstrate that CCFs and CALFs play a pivotal role in guaranteeing basic services for the elderly.

(2)

Calculating the base walk score incorporating distance decay

Travel distance decay characterizes the phenomenon wherein willingness to walk declines as travel distance increases. In this study, the decay function was calibrated to match the specific walking capabilities of the elderly [51,53] (Figure 3). Furthermore, the shortest walking distances extracted from online maps were converted into walking times for precise analysis.

(3)

Deriving the final score by applying attenuation indices for intersection density and block length

WS is adjusted according to the walking environment, and scores are standardized to a range of 0 to 100. After calculating the original WS based on distance, WS is adjusted according to street intersection density and average block length. Once WS for each community is obtained, it is divided by the maximum WS achievable with such facilities and then multiplied by 100 to standardize WS to a scale of 0–100 (

Table 5. Walkability evaluation criteria.

Walk Score	Description
90–100	Walker’s Paradise: Daily trips can be completely solved by walking.
70–89	Very Walkable: Most daily trips can be accomplished by walking.
50–69	Average Walkability: Some facilities are within walking distance.
25–49	Poor Walkability: There are fewer facilities within walking distance.
0–24	Car Dependent: Almost all trips depend on cars.

Source: walkscore.com/Walk Score Methodology (https://www.walkscore.com/methodology.shtml).

).

2.3.2. Analyzing the Non-linear Influencing Variables Based on XGBoost Model

(1)

Selection of influencing variables for the convenience of 15 min CLC

This study constructs a non-linear model to investigate the differential impacts of influencing variables on categorized versus overall facility convenience [54]. The convenience score, calculated via WS method, serves as the response variable. Drawing on the existing literature regarding 15 min CLC, influencing variables across five dimensions were incorporated into the XGBoost model [55,56] (

Table 6. Influencing variables affecting 15 min CLC and their description.

Dimension	Influencing Variable	Description of Influencing Variable
Socio-economy	Population density (PD)	Number of population in residential quarters (units)/ quarter area (km2)
	Proportion of elderly population over 65 years old (PE65)	Number of people over 65 years old in residential quarters/ total population
Construction scale	Residential plot ratio (RPR)	Total above-ground construction area/ net land area of residential quarters
	Building density of life circle (BDLC)	Total base area/total land area of life circle buildings
Functional property	Functional density (FD)	Number of POIs in the life circle (pieces)/ area of the life circle (km2)
	Functional mixing index (FMI)	Shannon diversity index of POI facilities within the life circle
Traffic condition	Road network density (RD)	Total road network length (km)/life circle area (km 2)
	Border vacuum (BV)	Nearest straight-line distance to a place that impedes walking
	Distance from the city center (DCC)	Straight-line distance from geometric centroid of residential quarters to city center
	Distance to the nearest park (DNP)	Straight-line distance from the geometric centroid of residential quarters to the nearest park
	Distance to the nearest metro (DNM)	Straight-line distance from geometric centroid of residential quarter to nearest subway station
Land use structure	Residential and non-residential mixing index (RNMI)	Degree of balance between residential land and non-residential land
	Proportion of industrial land (PIL)	Industrial land area/life circle area
	Proportion of parks and green spaces (PPGS)	Land area of parks and green spaces/life circle area

).

(2)

XGBoost model and interpretation methods

XGBoost (eXtreme Gradient Boosting) is an efficient ensemble learning algorithm based on the Gradient Boosting Decision Tree (GBDT) framework [57]. It optimizes the loss function by iteratively constructing multiple weak learners, typically CART regression trees [58]. Compared to traditional GBDT, XGBoost offers superior performance in both algorithm design and engineering implementation, making it the preferred choice for this study. The dataset was randomly partitioned into a training set (80%) and a testing set (20%). The final optimal parameter combination adopted includes 90 estimators, a learning rate of 0.05, a maximum tree depth of 7, and a gamma value of 0.05. To mitigate the inherent “black box” opacity of XGBoost 2.1.4, the SHAP (Shapley Additive Explanations) method was applied. By quantifying the contribution of each feature to the model output, SHAP enables interpretability analysis at both global and local levels. Globally, SHAP values reveal the average contribution and directionality of features; locally, they elucidate the formation mechanism of individual predictions by illustrating how specific features drive outcomes. This combination provides a robust explanatory basis for indicator optimization and causal inference.

3. Results

3.1. Results of Spatial Distribution Characteristics of Elderly Service Facilities in Shenhe District

KDE analysis reveals a distinct “west-dense, east-sparse” spatial pattern for elderly service facilities in Shenhe District, highlighting a structural lag in public service allocation within the eastern urban fringe (Figure 4). Beneath this overall pattern, facility types exhibit heterogeneous characteristics driven by differing service attributes and mechanisms. CCFs, serving as essential life support, demonstrate broad coverage across the district. Driven by universal demand, convenience stores and restaurants display a relatively uniform distribution. Conversely, large shopping malls are highly clustered in core areas such as Zhongjie and Qingnian Street, reflecting a clear commercial locational orientation. Due to scale requirements and rent sensitivity, vegetable markets exhibit significant service gaps, with blind spots identified in sub-districts including Wulihe, Nanta, Wanlian, and Maguanqiao. CALFs reflect strong planning-led characteristics. Fully government-funded venues, such as parks and elderly activity stations, show a balanced distribution, embodying the policy intent of equalizing public services. LSFs integrate commercial clustering with the spatial balance of public services. TFs align closely with the regional road network, featuring significantly higher density in the northwest compared to the south and east.

3.2. Analysis of Spatial Differentiation of Convenience in 15 min CLC

Ordinary Kriging interpolation was employed to visualize spatial differentiation in 15 min CLC in Shenhe District. This analysis generated accessibility results for both categorized and overall facilities based on walk score.

3.2.1. Convenience of Categorized Facilities

(1)

Convenience commercial facilities

Elderly residents enjoy high accessibility to convenience stores (walk score: 76.88) and restaurants (92.12) within a 15 min walk. In contrast, access to shopping malls is notably lower (32.23), with low-accessibility zones primarily concentrated in eastern sub-districts such as Quanyuan, Dongling, and Maguanqiao (Figure 5). Vegetable markets record the lowest convenience score (16.07), as the majority of residential locations fall beyond the 15 min walking threshold.

(2)

Life service facilities

The convenience of LSFs exhibits uneven characteristics. The spatial layout of express delivery points (16.29) presents significant challenges, with particularly poor convenience in the southwest hindering access for elderly residents. Conversely, telecommunication outlets (71.42) and neighborhood committees (68.54) demonstrate satisfactory convenience levels. While bank outlets (76.60) maintain a high average convenience across the district, they exhibit a distinct “polarized” spatial distribution. Police stations, characterized as low-frequency facilities, show a balanced distribution with a “multi-center” pattern along sub-district boundaries; although walking convenience is low (32.53), it remains sufficient for elderly needs (Figure 6).

(3)

Elderly care service facilities

The overall convenience of ECSFs is generally average, and there are obvious differences in their spatial distribution. Accessibility to elderly dining points (34.80) varies substantially. Residents along the southern riverside face the lowest walking convenience, whereas those near administrative boundaries benefit from cross-border service provision. Convenience for day care centers (41.26) and nursing homes (44.51) is moderate, yet a marked disparity persists between the eastern and western regions (Figure 7).

(4)

Medical and health facilities

Among the MAHFs, primary healthcare service points demonstrate excellent accessibility, yet hospital services in Shenhe District exhibit disparities in convenience between eastern and western regions. Pharmacies (88.29) and clinics (77.55) are highly accessible via walking, exhibiting minimal variation across life circles. Access to hospitals (50.34) is moderate but displays significant east–west polarization, with higher convenience predominantly found in western life circles (Figure 8).

(5)

Cultural and leisure facilities

Among the CALFs, the daily activity venues are more convenient, but the accessibility of professional education for the elderly and outdoor fitness venues is generally insufficient. While walking convenience to elderly activity stations (71.77) is relatively high, elderly activity centers (13.80) and universities for the elderly (11.08) sited at sub-district boundaries are difficult to access within 15 min, resulting in poor overall convenience. Generally, access to outdoor leisure services is deficient district-wide, ranked as follows: parks (18.27) > squares (11.18) > outdoor fitness venues (9.31) (Figure 9). High-value areas are concentrated in the southwest. In the northwest, park construction is constrained by tight urban land use; however, elderly residents in the Beizhan and Zhujianlu sub-districts enjoy better accessibility to squares, which moderately compensates for the deficiency in park services.

(6)

Transportation facilities

The accessibility of TFs varies significantly, with bus stops generally being easily walkable, while metros are limited in coverage. Convenience for metros (16.38) varies greatly, with high-accessibility zones distributed linearly along subway routes, leaving other areas underserved. In contrast, high convenience for bus stops (85.08) is balanced across the district, ensuring that most elderly residents can reach a stop within a 15 min walk (Figure 10).

A comprehensive comparison of the accessibility of facility categories in a 15 min CLC renders the following ranking: medical and health facilities (74.50) > transportation facilities (54.76) > convenience commercial facilities (46.17) > elderly care service facilities (38.85) > life service facilities (34.08) > cultural and leisure facilities (20.10) (Figure 11). Daily medical needs are adequately met via walking, and transportation convenience is moderate. The lower scores for CCFs and LSFs are primarily attributed to the poor distribution of specific high-weighted facilities, such as vegetable markets and express delivery points. Increasing the supply and optimizing the layout of these facilities would effectively enhance overall category scores. Conversely, CALFs and ECSFs require urgent improvement, as access is difficult across all life circles. This issue is particularly acute in the eastern part of Shenhe District, where large-scale industrial and green space land use generates more severe negative impacts compared to the west.

3.2.2. Convenience of Overall Facilities

The comprehensive evaluation reveals that walking accessibility for the elderly in Shenhe District is generally poor and spatially inequitable. With a district-wide average walk score of 40.52, only 96 life circles fall within the 50–70 range, indicating a widespread insufficiency in walking access to daily services. Spatial analysis highlights a distinct “west-high, east-low” macro-pattern, with convenience in the western region exhibiting a decreasing gradient from north to south.

At the sub-district level, convenience levels can be classified into three tiers. The higher tier (e.g., Huangcheng) scores near or above 50, offering relatively convenient walking access. The average tier scores between 40 and 50, reflecting moderate accessibility. The lower tier, concentrated primarily in the east, records significantly low scores, implying that residents are unable to rely on walking for basic needs. Furthermore, two spatial modes were identified: an atypical “edge-higher-than-center” feature in old core sub-districts, and a typical “center-high, periphery-low” pattern elsewhere.

3.3. Analysis of Non-linear Influencing Variables in 15 min CLC

Prior to modeling, the 14 selected variables were tested for multicollinearity. All variance inflation factor values remained below 5, confirming the absence of significant multicollinearity issues.

3.3.1. Relative Importance of Influencing Variables

The XGBoost model exhibited robust performance, with adjusted R² values ranging from 0.67 to 0.84 across overall and category-specific facility models. This indicates a high degree of model fit and explanatory power. The SHAP analysis method was employed to examine the impact of influencing variables on 15 min CLC convenience. Mean absolute SHAP values were calculated for 14 variables. The magnitude of these values reflects the relative importance of each variable in the model. To visualize these effects, Figure 12 converts SHAP values into relative contribution rates. A higher rate indicates a greater contribution of that indicator to the overall convenience of the life circle.

(1)

Convenience of overall facilities

Training results indicate significant disparities in influencing variables. The relative importance ranking is as follows: functional density (37.51%) > distance from the city center (21.91%) > population density (5.81%) > proportion of parks and green spaces (5.23%) > proportion of industrial land (5.06%) > road network density (4.73%). As the dominant variable, FD signifies facility abundance, directly enhancing accessibility. DCC and PD follow, influencing convenience indirectly by supporting service supply agglomeration. Land use variables (parks, industrial land) play a secondary, regulatory role. Notably, RD is also significant, suggesting that a well-developed road system expands effective service coverage. Within specific dimensions, impacts vary. Functional mixing index (2.90%) is considerably less influential than FD, suggesting that diversity offers limited benefits without sufficient facility quantity. Regarding construction scale, building density of life circle (2.90%) exerts a stronger influence than residential plot ratio (2.16%).

(2)

Comparative analysis of convenience of overall facilities and categorized facilities

Relative importance analysis revealed that certain variables maintained consistent contributions across different facility types, while others fluctuated. FD, DCC, RD, BDLC, and RPR demonstrated stable influence patterns. Specifically, FD (10.52–29.85%) and DCC (11.23–27.38%) remained the dominant drivers for most facilities. RD generally contributed between 5.95% and 12.91% (excluding CCFs and MAHFs). Conversely, variables related to construction scale had a limited impact, with BDLC and RPR contributing only 2.36–3.66% and 1.91–3.57%, respectively.

In contrast, variables such as FMI, DNM, DNP, and PIL showed increased explanatory power in category-specific models compared to the overall model. The contribution of FMI in MAHFs, ECSFs, CALFs, and CCFs increased by 4.17, 3.95, 4.04, and 2.01 percentage points, respectively. The contribution of DNM to CCFs, LSFs, ECSFs, and CALFs increased by 7.05, 5.71, 4.51, and 2.50 percentage points, respectively. DNP exhibited a non-linear influence (negative to positive) on CCFs and TFs, with contributions increasing by 3.98 and 2.63 percentage points. The influence of PIL on ECSFs and MAHFs strengthened by 2.18 and 3.30 percentage points, respectively.

However, the influence of PD and PPGS declined for most facility categories. PD retained high explanatory power only for CCFs and MAHFs, while its contribution dropped by 0.24–4.27 percentage points for other types. Similarly, the contribution of PPGS decreased universally across all facility categories, with a decline ranging from 1.07 to 3.74 percentage points.

3.3.2. Non-linear Response of Key Influencing Variables and Threshold Effects

Partial dependence plots elucidate the non-linear relationships between key influencing variables and convenience (Figure 13).

(1)

Functional density (FD)

Values exceeding 350 units/km² positively influence convenience. Beyond 535 units/km², impacts diverge: while CCFs and MAHFs stabilize, overall and TF convenience continue to rise (Figure 13A). Conversely, ECSFs exhibit a suppression effect, suggesting that high-density development may introduce noise or traffic externalities that degrade the quality of elderly care environments.

(2)

Distance from the city center (DCC)

DCC has an inhibitory influence on the convenience of most facilities (Figure 13B), among which the negative transformation critical points of OFs, CCFs, MAHFs, and TFs are 3.34 km, 2.85 km, 3.77 km, and 3.31 km, respectively. ECSFs and CALFs show an “inverted U-shaped” relationship, with a peak value of about 3 km, suggesting that this distance interval may correspond to the “optimal service radius” of facility layout. LSFs show a “U-shaped” relationship, which has a negative influence in the range of 2.76–7.37 km, and turns positive after exceeding 7.37 km, reflecting that the development of urban multi-center structure makes the outer suburbs form a self-sufficient secondary center.

(3)

Road network density (RD)

This variable shows obvious heterogeneity of facility categories (Figure 13C). Low RD (such as < 12.56 km/km²) inhibits the convenience of OFs, CCFs, and LSFs; after exceeding this threshold, some facility categories experienced brief fluctuations. When RD exceeds 40 km/km², its influence on most facilities tends to be stable. It is worth noting that ECSFs and CALFs have inhibitory effects after RD exceeds 25.62 km/km² and 22.57 km/km², respectively, reflecting that the high road network environment may interfere with quiet service facilities. MAHFs and TFs show an “inverted U-shaped” response, with positive intervals of 18.49–36.03 km/km² and 14.02–43.57 km/km², respectively.

(4)

Proportion of industrial land (PIL)

This variable presents the characteristics of “fluctuation suppression” for most facility categories (Figure 13D). When its proportion exceeds 0, it will inhibit OFs, CCFs, LSFs, ECSFs, and MAHFs. The anti-interference ability of CALFs and TFs is slightly stronger, and the inhibition critical points are 5% and 11%, respectively, which reflects the differentiated constraints of environmental externalities of industrial land on the layout of public service facilities.

4. Discussion

This study systematically evaluated the convenience and driving mechanisms of the 15 min CLC in Shenhe District from an aging perspective. It integrated spatial analysis with the XGBoost model. Methodologically, facility weights were determined based on survey data regarding the preferences of the elderly. Additionally, actual walking route data were used to calculate distances between residences and service facilities [59]. This approach reflects the real travel behavior of the elderly better than traditional GIS-based methods, such as the cumulative opportunity method, network analysis, and the two-step floating catchment area (2SFCA) method. Regarding results, the study reveals the spatial heterogeneity of facility supply. It also deepens the understanding of the complex relationship between the built environment and walking accessibility for the elderly.

4.1. Multi-Dimensional Patterns of 15 min CLC Convenience from an Aging Perspective

This study reveals a macro-spatial pattern in Shenhe District characterized by “west-dense, east-sparse” facility distribution and “west-high, east-low” convenience. Resonating with research on transitional Chinese cities and Western contexts [23,60,61], this finding indicates that urban fringe areas often suffer from a structural lag in public service allocation, which is fundamentally an issue of spatial equity. Specifically, the relatively high convenience of MAHFs [62] and TFs [63] reflects the preliminary success of policy interventions in basic public services. However, the severe deficiency in CALFs [17] and ECSFs [62] exposes a neglect within the current planning system regarding the elderly’s higher-level needs for spiritual comfort and long-term care. This “tiered disparity” suggests that current life circle construction remains in a nascent stage, primarily satisfying basic survival needs rather than the social participation goals advocated by “age-friendly cities”. Therefore, future planning interventions must transcend the mere “presence” of facilities to prioritize service “quality” and “inclusivity”, specifically addressing supply deficits in CALFs and ECSFs.

4.2. Role of Core Influencing Factors and Non-linear Thresholds

The study confirms that functional density (FD) is the most dominant driver of 15 min CLC convenience. Consistent with literature linking infrastructure sufficiency to increased walking probability [64,65], facility agglomeration forms the basis of service efficiency. Notably, however, FD exerts an inhibitory effect on ECSF convenience beyond specific thresholds. This implies that high-density development generates negative externalities echoing findings on the site selection dilemmas for elderly facilities in dense urban environments [66]. In addition to FD, distance from the city center (DCC) [67] and population density (PD) [23,63] serve as critical factors. The city center’s natural advantage, derived from historically accumulated functional density, verifies the “center–periphery” effect. Conversely, this suggests that cultivating high-density functional nodes in suburbs via planning is essential for mitigating spatial imbalances. While high PD attracts services and expands choices, its effect is largely indirect, supporting supply rather than constituting it; thus, its model contribution is slightly lower than that of FD.

Furthermore, the XGBoost model elucidates complex non-linear mechanisms, offering a crucial supplement to traditional linear assumptions. We found that proportion of parks and green spaces (PPGS) positively impacts convenience within specific ranges, revising the simplistic view of linear negative correlation [23]. Reasonable green space configuration optimizes slow-traffic networks and aggregates vitality. However, exceeding a critical value creates a “crowding-out effect” on facility land, thereby reducing convenience. Finally, contradicting the assumption that road network density (RD) is a simple positive variable [68], this study reveals significant type-specific heterogeneity. RD exhibits an “inverted-U” relationship with MAHFs and TFs, indicating that an optimal density range exists. Moreover, excessive RD inhibits ECSFs and CALFs, likely because the high-intensity development introduces safety risks and noise pollution, compromising the tranquil environments required by these facilities [69].

4.3. Limitations and Future Research

This study used POI data to identify facility types and quantities, effectively mapping their spatial distribution. We implemented online walking route planning between elderly service facilities and residential points. This approach focuses on location and number. Future research could build on this work by using methods such as street view semantic segmentation. This would allow the introduction of qualitative dimensions (e.g., service quality) and quantitative metrics (e.g., facility size) into the evaluation system, leading to a more comprehensive assessment of age-friendliness. Furthermore, the non-linear mechanisms revealed in this study provide a reference for identifying constraints on facility convenience in high-density, flat cities like Shenyang. The applicability of the specific threshold effects in other types of regions requires further verification. Despite these limitations, the analytical framework and core conclusions of this study offer significant theoretical and practical value.

5. Conclusions

Rapid population aging makes convenient urban living essential. In this context, 15 min CLC is a key strategy to ensure access to basic services within walking distance. This study focused on 510 life circles in Shenhe District, Shenyang. POI data and online walking routes were used to analyze the distribution of elderly services. The walk score method was employed to evaluate convenience. Using 14 influencing variables across five dimensions, convenience scores were predicted. Finally, the XGBoost model and SHAP method were applied to analyze the non-linear impacts of five dimensions, such as socio-economy, functional property, land use structure, etc. The main conclusions are as follows:

(1)

The spatial distribution of elderly services in Shenhe District shows a significant “west-dense, east-sparse” imbalance. This varies by facility type. The most critical shortages are found in vegetable markets, express delivery points, elderly dining points, universities for the elderly, and elderly activity centers.

(2)

Life circle construction faces challenges. Convenience follows a distinct “west-high, east-low” pattern. Convenience levels follow this hierarchy: medical and health facilities (MAHFs) > transportation facilities (TFs) > convenience commercial facilities (CCFs) > elderly care service facilities (ECSFs) > life service facilities (LSFs) > cultural and leisure facilities (CALFs). This indicates that medical and transport needs are met. However, support for social interaction and elderly care is severely insufficient.

(3)

Machine learning revealed key influencing mechanisms. Functional density and distance from the city center are the strongest predictors. Impacts vary significantly by facility type. For instance, functional mixing is crucial for MAHFs and CALFs, while industrial land has a negative impact on ECSFs. Threshold effects are evident. High FD is generally good but inhibits ECSFs. Road network density has an optimal range for all facilities. DCC shows an “inverted-U” effect on ECSFs. This suggests that planning strategies must be differentiated based on facility type, avoiding homogenized layouts.

Based on these findings, the following planning recommendations are proposed: First, implement precise supply strategies. New elderly care and cultural nodes should be developed in urban sub-centers. A comprehensive nursing system needs to be established combining home, community, and medical support. Additionally, activity centers and green leisure trails should be added near residential areas to promote healthy lifestyles. Second, utilize non-linear thresholds in planning. ECSFs should be located to avoid areas with excessive functional density. Optimal radii from the city center should be identified to balance service scale and cost. Regarding road networks, traffic calming measures are needed near ECSFs and CALFs. This ensures a safe walking environment. Furthermore, the critical limits of industrial land influence can provide a scientific basis for zoning regulations. Through these interventions, 15 min CLC can become a practical reality for the elderly.